WO2022007212A1

WO2022007212A1 - Software microservice combination optimization method for edge cloud system

Info

Publication number: WO2022007212A1
Application number: PCT/CN2020/119196
Authority: WO
Inventors: 徐小龙; 姜宇; 李洋
Original assignee: 南京邮电大学
Priority date: 2020-07-10
Filing date: 2020-09-30
Publication date: 2022-01-13
Also published as: CN112398899A; CN112398899B

Abstract

A software microservice combination optimization method for an edge cloud system which, by resetting a step factor and an attraction degree formula, fully leverages the probing ability of the step factor in the early stage of an algorithm and balances local search and global search. Microservices are classified according to the characteristics of the microservices, and services having the same inputs and outputs are classified into the same type, thereby reducing the number of combinations and making possible the enumeration of all combination schemes. Random initialization and random perturbation are replaced with cotangent initialization and cotangent perturbation in the combination algorithm to improve the initialization and updating of the microservice combination schemes and greatly enhance the diversity of a final combination scheme. Meanwhile, a service attribute value stored in a service registration center is hashed and stored as a root hash using the hash tree concept, thereby ensuring the security and reliability of each quality of service attribute value stored in the service registration center.

Description

A software microservice composition optimization method for edge cloud system

technical field

The invention relates to the technical field of information integration and software engineering application, in particular to a software microservice combination optimization method oriented to an edge cloud system.

Background technique

Since the advent of the "microservice architecture", the design of software application systems has shifted towards sets of independently deployable services. Compared with the traditional monolithic development style, microservices focus more on developing an independent application system by developing a set of small services, each of which runs in its own process. These services are built around business functionality and can be deployed independently through a fully automated deployment mechanism.

A monolithic application system is built as a single unit. As more and more application systems are deployed to the cloud, software changes are greatly restricted. Once a part of the application system is changed, the entire monolithic application system needs to be rebuilt and deployed. Obviously, monolithic application systems are increasingly difficult to maintain a good modular structure. For this limitation, the architecture of microservices can be well filled. In microservices, each service can provide a solid module boundary, even allowing different programming languages to be used to write different services. The "componentization" of microservices is characterized by decomposing software into many services, which can communicate through web service requests or remote procedure calls.

With the increasing number and complexity of current user needs, the problems that can be solved by a single service are becoming less and less, and the increase in complexity makes service composition more and more important, so that the combination of microservices can solve the problem. More problems, and in the concept of software engineering, the degree of aggregation is improved, the degree of coupling is reduced, it is more convenient to maintain on the basis of combination, and it is easier to add new functions and reduce original functions, as long as a separate microservice module After strict testing, the parameters in all aspects meet the requirements, we can use it with confidence, and can quickly locate the error after finding the error, the final service is a combination of tested microservices, and a single microservice module is correct. This makes the application of microservice composition more and more widely.

In the cloud computing environment, multiple microservices are combined into a software application system, that is, the system is decomposed into several services according to business functions. Everyone can use different programming languages and technical tools to develop their own microservice modules, and with the large increase in the number of published microservices, it becomes a difficult choice for software application systems to choose which microservices to use. This is caused by the widespread problem of service heterogeneity, which reduces the accuracy of service discovery, matching and selection, and the ability to interoperate between services, affects the effectiveness and correctness of composite services, and becomes a dynamic composite development. one of the bottlenecks.

The uncertainty of microservices is also relatively prominent. Uncertainty problems include whether the availability of microservices is uncertain. The service quality standards of microservices can be adopted and introduced into the evaluation of microservices. The quality of service (QoS) of microservices is dynamic, uncontrollable, and different. Of course, users have different requirements for microservice quality QoS. For different microservice application fields, the combination of microservices Schemas and associations are different. Uncertainty issues bring a variety of problems to microservice composition, which affect the effective design, development, reliability, availability, and quality of the system.

Microservices have been applied in the development of software application systems. The quality of service (QoS) problem is very critical to the successful application of microservices. How to provide microservices with QoS guarantees is a hot issue in the research and application of microservices. From the perspective of microservices, how to select appropriate microservices from a large number of microservices and optimize the combination so that the QoS of the service composition service can meet the requirements is an important issue in the research of microservice composition.

For today's microservices composition, most of them are aimed at the cloud computing environment. For cloud computing, in order to improve cost efficiency, multiple microservices are usually placed on the server at the same time, but with the increase of server resources (core, cache, bandwidth, etc.) With the increase and the diversity of microservices, the space for scheduling exploration is rapidly expanding. Cloud computing adopts microservices and container technology to increase the elasticity and scalability of cloud computing. Microservices are deployed in containers to add multiple copies of services to balance the load. . Edge computing is to sink tasks to the edge of the network. When the edge cloud also provides micro-services for users, in order to meet the increasingly complex application needs of users, the micro-service composition also needs to be applied in the edge environment, and the related service composition algorithms are also Improvements are needed to accommodate service mix in edge environments.

In the microservice composition, the matching problem of each service in a microservice composition is a complex composition optimization problem, that is, a large number of microservice collections are searched for a combination that satisfies a certain service quality and meets the needs of users. Solving this problem is not only time-consuming, but it is also difficult to find the best microservice combination solution. The results of the solution directly affect the quality and cost of the microservice combination. Especially in the edge environment, each edge server is scattered in various locations. Combining microservices deployed on edge servers requires high costs. In response to this problem, the use of intelligent optimization algorithms to solve the microservice composition optimization problem is the current mainstream idea. This optimizes the microservice composition to a certain extent, but there are still the following shortcomings:

1) As the number of microservices increases, the amount of computation increases exponentially, so the efficiency of solving optimization problems is low;

2) In the process of searching for the optimal solution, the random search strategy cannot guarantee the diversity of the final solution, and a new search strategy is required;

3) When solving the microservice composition optimization problem, it is necessary to consider not only the selection of microservices, but also the logical relationship between microservices;

4) In the edge environment, due to the scattered locations of edge servers, microservices are also in a scattered state, and it is difficult to combine microservices. The combination optimization algorithm needs to consider the characteristics of the edge environment;

5) The data stored in the service registry may be subject to malicious attacks at any time. The most important thing is to tamper with the service attribute value. It is also a very important issue to ensure the safety and reliability of each service quality attribute value stored in the service registry.

SUMMARY OF THE INVENTION

Purpose of the invention: The purpose of the present invention is to provide a software microservice combination optimization method oriented to edge cloud systems. The QoS service quality of the microservice combination is guaranteed to be optimal, and the efficiency of the microservice combination is further improved, and the reliability of the service quality of the provided microservice is also guaranteed.

Technical solution: A method for optimizing software micro-service combination for edge cloud system described in the present invention includes the following steps:

S1: The user submits his own task, which includes the corresponding task input and the expected output result. The task is submitted to the edge cloud, received by the edge orchestration manager, and deployed in each edge server through the registration center. Microservices information, and choose a combination of services;

S2: Introduce the predatory search strategy, and use the predatory search strategy to obtain a feasible service chain globally, that is, to search for a service chain that meets the requirements of the user's microservice composition;

S3: The length of the feasible service chain is m, and the candidate service corresponding to each microservice on the feasible service chain is searched, that is, the microservices with the same input set and output set as the microservices on the feasible service chain are put into the same service class , get m candidate service classes;

S4: In the candidate service class, use a cotangent sequence with chaotic properties to select a candidate service from each candidate service class in turn to form a micro-service combination. The micro-service combination is mapped to a driver individual, according to the previous driver The individual performs chaotic initialization n-1 times, which obtains n light-driving individuals;

S5: Evaluate the n microservice composition schemes according to the individualized fitness function, find the optimal microservice composition scheme among the n microservice composition schemes, and judge whether the corresponding fitness function value reaches the theoretical maximum If yes, then the microservice composition scheme is the global optimal microservice composition scheme; otherwise, execute S6, when the final optimal microservice composition scheme is found or the number of iterations reaches the upper limit, stop, and output the maximum value when stopped. Excellent microservice composition scheme;

S6: Update the micro-service combination scheme of 20% of the light-driving individuals with the chaotic disturbance rules, and these 20% of the light-driving individuals are to select 10% of the light-driving individuals with the best fitness value and 10% of the fitness value in the current iteration For the worst optical drive individuals, the rest of the optical drive individuals are still calculated according to the position transformation formula of the original combination algorithm. 20% of the optical drive individuals need to update their positions, and then perform a local search. After the update is completed, return to S5;

S7: The global optimal solution based on the current service chain is obtained through the combined algorithm, or the maximum number of iterations is reached, and the corresponding service chain is the current optimal service chain. After the microservice combination is successfully selected, the user's task is executed, and then the edge orchestration manager will return the execution result to the user, update the QoS attribute values in the corresponding microservice, and save the updated result in the Microservice registry;

S8: The edge orchestration manager will save the QoS attribute values of the microservices registered in the service registry in the service registry, and use the hash tree concept to hash the QoS attribute values of the microservices in pairs, similar to a tree structure Finally get a hash root. The edge orchestration manager encrypts the hash value with the private key, and then broadcasts it to each edge server. The edge server decrypts the broadcast information with the stored public key and saves the hash value. The edge server will periodically ask the registry to see if the hash root stored in the registry is consistent with the locally stored hash value. If it is inconsistent, it is judged that the service QoS attribute value stored by the service registry has been tampered with, and the attribute value of which service has been modified can be quickly located by using the properties of the hash tree, and the specific edge server can also be located.

Preferably, the specific process of the S2 is as follows:

S2.1: Set the searchSet collection to be empty, search all microservices, add the microservices that meet the user's microservice composition requirements to the service chain, and store the input collection of the microservices in the searchSet collection;

S2.2: Before adding a microservice that meets the user's microservice composition requirements to the service chain, determine whether there is an input set of the microservice in the searchSet collection, if not, add the microservice to the service chain, otherwise, do not join ;

S2.3: After the current search is over, if no feasible service chain is obtained, the searchSet collection will be cleared, and S2.1-S2.2 will be repeated in the microservices that have not been searched until a feasible service chain is found or the maximum search is reached. times, stop searching.

Preferably, the specific process of the S4 is as follows:

S4.1: The length of the feasible service chain is m, j=0,...,m-1, and the jth microservice has k _j candidate services;

S4.2: Determine the order of magnitude of _{k j}

According to the cotangent sequence value corresponding to the jth microservice

Truncate the cotangent sequence value with φ _j bits after the decimal point as an integer value u _i,j ,

i=0,...,n-1, n is the total number of drive individuals;

S4.3: _{Take the remainder of u i,j} to k _{j to} generate a chaotic value ξ _i,j , the chaotic value corresponds to a candidate service in the candidate service class corresponding to the jth microservice, and this value corresponds to the If a certain service in the candidate services corresponds, the corresponding candidate service is selected. Similarly, after all services in the service chain perform this operation, the _{initialization of the position X i of} the i-th optical drive individual is completed.

Preferably, the calculation formula of the fitness function value in the S5 is:

Among them, f represents the fitness value, R represents the product of the reliability of each microservice on the service chain, T represents the sum of the response times of each microservice on the service chain, C represents the sum of the execution costs of each microservice on the service chain, and A represents the service The product of the availability of each microservice on the chain, α represents the weight of R, β represents the weight of A, γ represents the weight of T, η represents the weight of C, and α+β+γ+η=1.

Preferably, the specific process of the S6 is:

S6.1: The screened 20% of the light-driving individuals need to update their positions according to the chaotic sequence. The formula is as follows:

In the above formula

Indicates the position of the driver i in the jth dimension of the t iterations, that is, the selection scheme of the jth component of the service chain. The position update also needs to be calculated according to the cotangent sequence. At the t+1th iteration, the jth dimension position of the ith driver individual

The update is obtained iteratively according to the cotangent sequence obtained at the tth time, where h represents the dimension of the individual driver, that is, the length of the service chain, and n represents the maximum number of iterations. The updated cotangent function value needs to calculate the corresponding candidate service according to S4;

S6.2: 20% of the CD-ROM individuals conduct a local search to find their own optimal solution for local micro-service combination. Here, a chaos count, chaoNum, is set to limit the number of comparisons. When the CD-ROM individuals find the local optimum in the local search Or if this number of chaos is exceeded, the search can be ended;

S6.3: The light-driving individuals participating in chaotic disturbance need to compare the locally optimal combination scheme found by themselves with the historical global optimal combination scheme. If it is better than the historical global optimal combination scheme, update the historical optimal combination scheme with this scheme as the new fBest, otherwise continue to S5.

Beneficial effect: the present invention adopts the above technical scheme compared with the prior art, has the following technical effects:

1. The microservice composition optimization problem is a complex NP-Hard problem with a polynomial time complexity. With the expansion of the problem scale, it will lead to composition explosion, and it is unrealistic to completely traverse all composition schemes. The present invention classifies the microservices by utilizing the characteristics of the microservices, and divides the services with the same input and output into the same class, thereby reducing the combination reduction and making it possible to completely enumerate all combination schemes.

2. The present invention introduces the cotangent sequence method with chaotic properties. In the combined initialization stage, the cotangent sequence is used to initialize the position of the driver individual, and the ergodicity, regularity and pseudo-randomness of the cotangent sequence are used to make up for the original The random search strategy with the algorithm improves the overall search efficiency of the algorithm; in the chaotic disturbance stage, a new set of disturbance rules is used to fully disrupt some of the drive individuals, so that the algorithm is not easy to fall into local optimum and has global search ability .

3. The present invention improves the step size factor and the attraction formula of the algorithm, so that the random item can exert its effective detection ability, balance the local search and the global search, and also prevent the attraction between population individuals due to excessive distance. When it is 0, the optimization ability is improved.

4. Based on the edge computing environment, deploy microservices and application service composition algorithms in the edge cloud, provide services to users at the edge, reduce transmission delay, and improve user experience; the concept of hash tree is introduced to improve the QoS of services. The attribute values are hashed layer by layer and stored in each edge server to ensure the reliability of the provided microservices.

Description of drawings

Figure 1 is a simulation diagram of the optimization of the microservice composition scheme;

Figure 2 is a schematic diagram of falling into a local optimum;

FIG. 3 is a schematic diagram of a disruptive service combination scheme;

Figure 4 is a logical structure diagram of microservice composition;

Figure 5 is the service chain and candidate services;

Figure 6 is an edge cloud system architecture diagram;

Figure 7 is a schematic diagram of a hash tree;

FIG. 8 is a flow chart of a method for optimizing software micro-service composition for an edge cloud system according to the present invention.

detailed description

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but not to be construed as a limitation of the present invention.

The invention is an optimization method for software micro-service combination oriented to edge cloud system, which ensures that the QoS service quality of the micro-service combination provided for users is optimal under the premise of satisfying user service requirements in an edge environment, and further improves the Efficiency of microservice composition and increased diversity of microservice composition. In the traditional cloud computing environment, intelligent optimization algorithms are often used to solve microservice composition optimization problems, such as particle swarm optimization algorithm, genetic algorithm, simulated annealing algorithm, etc. Aiming at the micro-service combination optimization problem, the present invention also adopts an intelligent optimization algorithm to be improved to be suitable for solving the micro-service combination optimization problem. The intelligent optimization algorithm is a group-based random search algorithm. Each individual in the group is a candidate solution to the corresponding problem, and the optimal position is searched through group intelligence. Because of its simple concept, easy implementation, and good performance The advantages of optimization performance and other advantages are widely used to solve various optimization problems. Aiming at the optimization problem of micro-service combination, the invention improves the intelligent optimization algorithm. By introducing chaos theory in the initialization and update of the population, and optimizing the algorithm itself, a software micro-system oriented to edge cloud system is proposed for the edge environment. Service Portfolio Optimization Methods.

In the present invention, each combination of micro-services is represented as a light-driving individual. When the population is initialized, the position of each light-driving individual needs to be initialized, thereby determining the distance of each individual from the optimal solution. As shown in Fig. 1, there is an optimal solution, that is, an optimal combination scheme, and X ₂ ,...,X _i ,...,X _n-1, etc., represent other combination schemes. Here, X _j (j=0, 1, 2,..., n-1) is used to represent a certain microservice composition scheme. Among X ₀ , X ₁ ,...,X _i ,...,X _n-1 , each The vectors all represent a microservice composition scheme. All microservice composition schemes are compared, and the globally optimal microservice composition scheme is selected.

Suppose X _i micro program service composition from the optimal combination of micro programs and services Recently, the set X _i is the current global optimum combination of micro program service, and let the global optimal solutions for the micro-service composition X ^g. In the process of optimization, each microservice combination solution will approach a combination solution that is better than its own. However, in order to get close to the real global optimal combination scheme, the current optimal combination scheme will make random changes to find out whether there is a better combination scheme than itself. In order to avoid the loss of the optimal solution caused by random disturbance, the method of particle swarm optimization (PSO) historical global optimal solution is introduced here, and the optimal solution of the cumulative iteration is saved. Each micro-combination scheme will be attracted by a better solution, gradually approach the better solution, and update its position. The update formula is shown in formula (1):

Among them, the first item X _i (t) represents the position of the driver i in the t-th iteration, the second item X _j (t) represents the position of the driver j in the t-th iteration, and the third item represents the random term, generally ε _{i is a} random number, α∈[0.1].

The above optimization model considers the situation where there is only one globally optimal microservice combination scheme. Usually, the optimization problem is a multi-peak optimization problem, that is, there are multiple extreme values and multiple optimal values. Therefore, when the traditional combinatorial optimization algorithm solves the multi-peak optimization problem, due to the fast convergence speed, it is easy to fall into the local optimum in the calculation process, resulting in premature population. The situation of falling into a local optimum is shown in Figure 2. In the figure, there is an optimal microservice composition scheme and two local optimal microservice composition schemes. The current microservice composition scheme is closest to the local optimal microservice composition scheme A, and then all the microservice composition schemes are approaching, falling into the local optimal microservice composition scheme. That is, the state of the microservice composition scheme before the update, and all the microservice composition schemes are close to the local optimal microservice composition scheme A.

The present invention uses cotangent disturbance to improve the update of the microservice composition scheme, that is, the cotangent disturbance makes the microservice composition scheme fully disturb the position and trend of the microservice composition scheme during the optimization process, thereby greatly reducing the risk of falling into The probability of the local optimal value; as shown in Figure 3, it is the state of the microservice composition scheme after the update. After the update, the position and trend of the microservice composition scheme are fully disturbed, and the microservice composition scheme falls into the local optimum. For the multi-peak optimization problem, the present invention can search for as many globally optimal micro-service combination schemes as possible, ensuring the diversity of the combination schemes, and at the same time, using the concept of hash tree, the service attribute values stored in the service registry are analyzed. Perform hashing, and the obtained hash tree root value can determine whether the data is legal.

For the convenience of understanding the technical solutions of the present invention, some concepts are defined below:

Definition 1. Drive individual Each microservice combination is represented as a drive individual, and the position change of the drive individual corresponds to the change of the microservice combination scheme.

Definition 2 Chaos is to transform the variables from the original chaotic space to the solution space according to certain rules, and use the regularity, randomness and ergodicity of chaos itself to conduct chaotic search, jump out of the local optimum, and find the global optimal solution. .

Definition 3 The chaotic search method is to obtain the random motion state points with the characteristics of pseudo-randomness, ergodicity and regularity from the deterministic equation.

In the present invention, the cotangent formula with chaotic properties is used as the chaotic search method, and its formula is shown in formula (2):

_{a n + 1 = cot (a} n) (2)

_{a n} in formula (2) is a value of the cotangent sequence. When an initial value of the cotangent sequence is given, it needs to satisfy a ₀ ∈(0,π), and iterating the initial value through the cotangent sequence formula can generate A pseudorandom sequence. The "cotangent formula" is a classic example of the butterfly effect. For example, take the initial values of three cotangents as 1, 1.0001, and 1.00001 respectively, and use the cotangent formula to iterate the initial values respectively. Each item of the three series is the cotangent of the previous item. When the 10th item is calculated, The three series began to form huge divergences. This is the chaotic sequence. After enough items, the resulting numbers are random, chaotic, and traversed.

Definition 4. Service class (S _j ) S _j is a set with the same service function, and the services with the same input set and output combination are classified into the same service class, expressed as S={S ₀ , S ₁ ,...,S _m-1 }, where S _j represents the j-th type of service.

Definition 5 Candidate Services

is the basic logical unit that constitutes the service composition, and the candidate service of the jth service class is

The input set corresponding to its service class S _{j is}

The output set is

Each candidate service contains a QoS vector

The vector contains 4 parameters:

①Execution Cost (c): The execution cost c of a service refers to the cost to be paid as a service user to request the execution of the service.

②Service execution time (Response Time, t): The execution time t of the service is equal to the time between the time when the request is sent and the time when the result is received.

③Availability (a): The availability of a service is the probability that the service can be used, and its calculation formula is:

a=T _λ /λ (3)

where λ is a constant set according to the type of service, and T _λ is the time the service is available within time λ.

④Reliability (r): The reliability r of a service is the probability that a request is correctly responded within the maximum expected time, and its expression is:

r=N _s /N _t (4)

where N _s represents the number of times the service was successfully invoked _{during the observation time, and N t} represents the total number of times the service was invoked during the observation time.

Definition 6 The service chain (L _c ) selects the appropriate service from the candidate services, and according to its input and output logic, combines it into a sequential directed graph in logical order, which is the service chain, as shown in Figure 4.

The optimization model of the microservice composition oriented by the present invention is as follows:

(1) Suppose there are m service classes in total, that is, S={S ₀ , S ₁ ,...,S _m-1 };

(2) Each service class consists of several candidate services, namely

where k _j _{represents a total of k j} candidate services in the j-th service class;

(3) Each service has attribute parameters describing its quality;

(4) Suppose there is a service chain L _c =s ₀ →s ₁ →...→s _j →...→s _m-1 (s _j belongs to the service class S _j );

(5) Service portfolio to improve service quality as much as possible.

st getOutput(s _j )==getInput(s _j+1 )s _j ∈L _c

α+β+γ+η=1

Among them, s _j represents a service on the service chain, and the getInput() and getOutput() functions are to obtain the input set and output set of the corresponding service. α, β, γ, and η represent the weight of each QoS attribute, respectively, R, T, C, and A represent the product or sum of each QoS attribute of the service chain, respectively, where R and A are reliability and availability, respectively, multiple service combinations The total reliability is the product of the reliability of each service, and the availability can be obtained in the same way; while T and C represent the execution time and execution cost respectively, and the total execution time of multiple services is the sum of the time of multiple services. The total cost of execution can be reasonably obtained. However, because different QoS attributes have different units and different orders of magnitude, it is difficult to directly compare all QoS attributes, and the greater the time and cost attribute values, the lower the QoS evaluation. On the contrary, availability and reliability are proportional to QoS values. Therefore, it is necessary to unify the dimensions of the QoS attributes and perform normalization processing.

Among them, Q _d,k represents the d-th candidate service,

represents the sum of the kth quality attributes of _{m d} candidate services in the candidate service set. As mentioned in the previous paragraph, the attribute proportional to the QoS value is calculated according to formula (6), and the attribute that is inversely proportional is calculated according to formula (7).

According to the above definition, the present invention is divided into three steps: first, a feasible service chain is globally searched, and candidate services of each service on the service chain are found; secondly, the driver individual is chaotically initialized by the cotangent sequence; The fitness function value of the individual disc drives is used to evaluate the pros and cons of each disc drive individual, the optimal disc drive individual is selected as the object to be learned, and other disc drive individuals are updated.

First, a feasible service chain is obtained globally using the predator search strategy. All microservice data are stored in xml files, such as

<servname>FoodService</servname>

<input>USFoodInfo</input>

<output>HotelInfo</output>

</webservice>

Among them, servname represents the name of the service; time, availability, rel, and cost are response time, availability, reliability, and execution cost, respectively. input represents the input set of the service; output represents the output set. In order to be more in line with the edge environment, binding the microservice to the edge server means that the microservice is deployed in the edge server.

In the process of searching the service chain, in order to ensure that there are no invalid services in the service chain (that is, services that have no influence on the final output result), a search set is added, which is represented by the variable searchSet. In a search process, the input set of the microservice on the service chain is stored in this variable. Before the microservice is added to the service chain, it is necessary to determine whether the microservice input set already exists in the searchSet. If it does not exist, and the requirements are met , then the microservice can be added to the service chain, otherwise, it cannot be; when the search is over and no feasible service chain is found, the search set will be cleared, and the microservices that have not been searched will be searched until one is found. A viable service chain, or after the maximum number of searches, stop searching.

Definition 7 Cotangent initialization is to map the pseudorandom sequence value generated by the cotangent sequence to the initial value of the position of the population individual through certain mapping rules.

After predation search, a feasible service chain has been searched globally. Before chaotic initialization, the parameters required in chaotic initialization are explained first:

(1) Search for a globally feasible service chain according to the predator search strategy, assuming that the service chain is composed of m services;

(2) According to the services on the service chain, find the candidate services of the service chain, each service on the service chain corresponds to a type of candidate service, and let each type be S _j ={s _j,0 ,s _j,1 , ...}(j=0,1,...,m-1), the service chain and its corresponding candidate services are shown in Figure 5;

(3) Divide the microservices with the same function into one category (the present invention classifies the same input set and output set into the same category); through the classification, the number of combinations is reduced, so that it is possible to completely enumerate all the combination schemes;

(4) Suppose the number of each candidate service item is k _j (j=0,1,...,m-1), so

(5) According to the two points (1) and (2), it is assumed that the position X _i = (S ₀ , S ₁ , . . . , S _m-1 ) of the light-driving individual.

The microservice composition optimization problem is uncertain. The uncertainty here mainly means that the optimal moderate function value for evaluating the merits of the composition scheme is not known in advance, so we can only find a better solution as much as possible; and different service chains The number of candidate services corresponding to different services is also uncertain. Corresponding to different service chains, the cotangent sequence required for initialization will be different. Therefore, this paper proposes a dynamic cotangent sequence method to initialize the individual drive.

Both the cotangent sequence and the Logic sequence are chaotic, the Logic sequence is a sequence generated by the Logic mapping formula, that is, a linear function iteration, and the cotangent sequence is a sequence generated by the cotangent function iteration. Although the calculation time of the cotangent function is slightly larger than that of the linear function, the sequence generated by the cotangent function is more dynamic and ergodic. Therefore, the present invention employs a cotangent function.

In order to meet the application of micro-service combination, the present invention designs a dynamic chaotic sequence, and the main steps are as follows:

(1) Assume that the type of microservices on the service chain is m, and the number of candidate services corresponding to each service is k _j ;

(2) m random double type values between 0 and π as the initial value of the sequence

(3) substituting equation _{a n + 1 = cot (a} n) are sequentially calculated by the cotangent of sequence values;

(4) The order of magnitude of the candidate service of a certain type of service determines the number of decimal places to intercept the cotangent sequence; for example, if there are 70 service classes in the j-th type, take two decimal places, and use the intercepted number to calculate the remainder of 70, and the generated The number between 0 and 59 corresponds to a certain candidate service of this class.

Definition 8. The fitness function value is the value of the fitness function, which is a quantitative index used to evaluate the pros and cons of the microservice combination scheme.

The service composition of the present invention includes four QoS parameters, namely reliability, response time, execution cost and availability. Among them, reliability and availability are a probability, and their values are between 0 and 1. Response time and execution cost are not a probability. As shown in the above formula (7), they need to be unified and normalized. The strategy adopted in this paper is: The characteristics of each attribute of QoS, such as response time and execution cost, should be inversely proportional to the fitness value, normalized according to the above formula (7), reliability and availability are proportional to the fitness value, and normalized according to the above formula (6) Unification, the fitness function is:

Among them, f represents the fitness value, R represents the product of the reliability of each microservice on the service chain, T represents the sum of the response times of each microservice on the service chain, C represents the sum of the execution costs of each microservice on the service chain, and A represents the service The product of the availability of each microservice on the chain. α represents the weight of R, β represents the weight of A, γ represents the weight of T, η represents the weight of C, and α+β+γ+η=1.

The present invention is based on a micro-service scenario. If the method of the present invention is applied to other orderly combination optimization scenarios, the fitness function can be modified to be an indicator for evaluating the pros and cons of individual drivers in a specific scenario.

Definition 9 Chaos disturbance means that the microservice composition scheme introduces a cotangent disturbance method in the update process, fully disrupts the microservice composition scheme, and traverses the search space as much as possible.

When an individual driver falls into a local optimum, the individual driver should be allowed to jump out of the local optimum as much as possible. By adopting chaotic disturbance, the position of the light-driving individual is re-initialized according to the chaotic sequence. After re-initialization, it is necessary to search locally to find the local optimal value. If the optimal value found by the local search is better than the historical global optimal combination scheme, the replacement will be carried out, and the current chaotic disturbance will end, otherwise the chaotic disturbance needs to be repeated until a new global optimal solution is found or the set chaotic disturbance is reached. number of iterations.

Definition 10 Brightness is the fitness value corresponding to the location (combination scheme) of each light-driving individual.

Here, the brightness is defined as the fitness value of the light-driving individual, and the light-driving individual with low brightness tends to move toward the light-driving individual with high brightness. The formula is as follows:

In the formula, I _i represents the absolute brightness of the light-driving individual i, which is directly determined by the fitness value. γ is the light absorption coefficient, which is generally set as a constant. The effect of the light absorption coefficient is that the presence of air and other substances will lead to the weakening of light. Introduced into the combination optimization, that is, the better solution with a longer distance will weaken the attraction as the distance increases. degree, which is also reflected in the definition of attraction degree; r _ij is the Euclidean distance between the light-driving individual i and the light-driving individual j. Then the distance between the drive individuals i and j can be expressed as:

In formula (10), X _i and X _j are the spatial positions of the light-driving individual i and the light-driving individual j, respectively, _{xi, d} represent the d-dimensional component of the light-driving individual i, and D represents the dimension.

Definition 11 Attraction is the attraction between individuals, which depends on the influence of distance.

The attractiveness between the drive individuals in this algorithm is as follows:

Among them, β ₀ represents the maximum attractiveness, that is, the attractiveness of the driving individual at a distance of 0 (r=0). The minimum attractiveness _βmin is to ensure that the attractiveness between individuals is not 0. After analysis, if β _min = 0, when the distance between individuals approaches infinity, the attractiveness β approaches 0. Obviously, this will lead to random walks of population individuals due to the large distance between individuals, thereby losing traction on other individuals and losing the ability to seek optimization.

After adding β _min , take β _min =0.25, which can prevent the lack of attractive traction between the driving individuals. Even if the distance is too great,

approaching 0, but due to _{the existence of β min} , the attractiveness is not 0, and _{the introduction of β min} can solve the problem of lack of attractiveness when the distance between population individuals is large in the high-level combinatorial optimization problem, and is suitable for solving micro-combination problems.

Definition 12 The step size factor is the factor that affects the position change in the position change. Generally, α is used to represent the step size factor, and α∈[0.1].

In an ideal situation, the population of the algorithm will gradually gather during the convergence process, and eventually converge to a point. For the individual X _i , we can get:

lim _t→∞ X _i (t+1)=lim _t→∞ X _i (t) (12)

According to the position formula (1), it can be obtained,

lim _t→∞ α=0 (13)

It can be seen from equation (13) that when the algorithm converges, the step size factor α tends to 0.

As a random item in the general position transformation formula, the step size factor α takes a fixed value of 0.5, and ε _i in _{αε i} takes a random number. Obviously, the random item depends to a large extent on ε _i , then this item becomes a completely random item , obviously it is difficult to exert the detection ability and search ability of its adjustment algorithm during the operation of the algorithm. In order to play the role of the step size factor in the algorithm to a large extent, the present invention sets α, that is, in the early stage of the algorithm, the random item needs to have a large random step size to exert an effective detection ability. Reduce the step size to exert its effective local search ability, the formula is as follows:

α(t+1)=α(t)·exp(-0.5·t/T _max ) (14)

In the above formula, T _max represents the maximum number of iterations, and t is the current number of iterations. As the number of iterations increases, α gradually decreases. The initial value α _{0 of} α is generally set to be 0.5.

Definition 13 An edge cloud is a cluster composed of multiple edge servers in edge computing.

Traditional microservices are deployed in the cloud computing environment, so most of the microservice combinations are also for the cloud computing environment. In edge computing, tasks are sunk to the edge of the network, and services are provided to users nearby. Among them, major cloud manufacturers gradually provide some microservices to users at the edge to meet the needs of users nearby. In order to improve the user's service experience, the relevant combination optimization algorithm also needs to be improved according to the characteristics of edge computing to adapt to the service combination in the edge cloud.

In the edge cloud environment, each edge server is deployed close to the terminal. As shown in Figure 7, in the entire edge cloud, there is a server that acts as the edge orchestration manager to manage the registration and management of each microservice in the edge cloud in a unified manner. After the user's device is connected to the edge cloud through the base station, the task is uploaded to the edge orchestration manager, and the orchestrator selects the appropriate microservice combination scheme for the user in the microservice registration center, completes the user's task, and returns the result to end users.

Definition 14 Hash tree The hash tree structure is similar to a binary tree, and the value on the leaf node is usually the hash value of the data block, not the value on the leaf node.

When constructing a hash tree, the first step is to calculate a hash value for the data. Usually, a hash algorithm such as SHA-256 is used. But if the only protection against data is not intentional damage or tampering, some less secure but more efficient checksum algorithms, such as CRC, can be used instead. Then pair the hash values calculated by the data in pairs (if it is an odd number, the last one is paired with itself), calculate the hash of the previous layer, and repeat this step until the root hash value is calculated. If any underlying data block changes, it will eventually propagate to the root hash. If the root hash is inconsistent, the data causing the inconsistency can be quickly located through the hash tree. The hash tree can also be used to quickly compare data and quickly locate inconsistent data, as shown in Figure 8.

In the present invention, the edge orchestration manager will save the QoS attribute values of the microservices registered in the service registration center in the service registration center, and use the concept of hash tree to hash the QoS attribute values of the microservices in pairs, and finally obtain a hash root. The edge orchestration manager encrypts the hash value with the private key, and then broadcasts it to each edge server, and the edge server saves the hash value with the saved public key. The edge server will periodically ask the registry to see if the hash root stored in the registry is consistent with the locally stored hash value. If it is inconsistent, it is judged that the service QoS attribute value saved by the service registry has been tampered with, and the attribute value of which service has been modified can be quickly located by using the properties of the hash tree, so as to ensure the reliability of the microservice QoS attribute value. .

The present invention takes microservice composition as an example, and the quality in the service quality determination problem includes service execution time, execution cost, reliability and availability. The production flow chart of the microservice combination scheme of the present invention is shown in FIG. 8 . The specific operation steps are as follows:

S1: The user submits his own task, and the task includes the corresponding task input and the expected output result. The task is submitted to the edge cloud, received by the edge orchestration manager, and obtains the information of the microservices deployed in each edge server through the registration center, and selects the combined service.

S2: First introduce the predatory search strategy, and use the predatory search strategy to obtain a feasible service chain globally, that is, to search for a service chain that meets the requirements of the user's microservice composition. Its specific description is as follows:

① In a search process, the input set of the microservices on the service chain is stored in the searchSet collection variable;

②Before the microservice is added to the service chain, it is necessary to determine whether the microservice input set already exists in the searchSet collection. If it does not exist and meets the requirements, the microservice can be added to the service chain, otherwise, it cannot be;

③ When the search is over, and no feasible service chain is found, the search set is cleared, and the microservices that have not been searched are searched until a feasible service chain is found, or the search is stopped after the maximum number of searches.

S3: Assuming that the length of the service chain is m, search the candidate services corresponding to each service on the feasible service chain in S2, and cluster the candidate sets of each abstract task, that is, the microservices with the same input set and output set are grouped together. As one class, put them into the same service class, so as to reduce the number of combinations and improve the search efficiency. According to the length of the service chain, m candidate service classes can be obtained, that is, S={S ₀ , S ₁ ,...,S _m-1 }.

S4: Then, in the initialization of the combined algorithm, the chaotic search method is introduced to replace the random initialization, and the chaotic sequence is used in the candidate service to select a candidate service from each candidate service in turn to form a feasible service chain, and each service chain corresponds to a driver. individual, perform the initialization operation multiple times to generate multiple optical drive individuals; each service on the service chain has k _j candidate services, and it is assumed here that the i-th optical drive individual is initialized. The specific steps are as follows:

① First determine the order of magnitude of _{k j} _{φ j}

According to the cotangent sequence value corresponding to the service

_{Truncate φ j} bits after the decimal point as an integer value u _0,j , whose range is

(excluding

);

②Then take the remainder of _{u 0,j} to k _{j to} _{generate a chaotic value between 0～k j} (excluding k _j ), here the chaotic value is set as ξ _0,j , which is the candidate corresponding to the service If one of the services corresponds to one of the services, the corresponding candidate service is selected. Likewise, after all the services in the service chain perform this operation, the _{initialization of the position X i of} the i-th optical drive individual is completed.

Assuming that there are n drive individuals in total, after the initialization of the first drive individual is completed, the second drive individual needs to be initialized according to the first drive individual, that is, using the cotangent function, the first drive individual The cotangent result is used as a variable to calculate its own cotangent value, and the remaining optical drive individuals proceed in turn. Then it takes n-1 times of the above process, and an n×m cotangent sequence matrix C can be obtained. As shown in formula (15), the cotangent sequence matrix is obtained by intercepting and taking the remainder operation to obtain a one-to-one correspondence with the individual driving light. matrix. The specific steps are as follows:

①Iterate n times, that is, generate an n×m cotangent sequence matrix, as shown in equation (15). Starting from the second drive individual, it is obtained according to the cotangent of the previous drive individual, as shown in formula (16):

(2) Determine the number of digits after the decimal point of the cotangent sequence matrix according to the magnitude of the corresponding candidate service quantity φ _j _{, and take the remainder of the truncated value to k j to} obtain the chaos matrix Φ, as shown in Equation (17).

During initialization, the initial value of each parameter is shown in equation (18).

S5: Evaluate the n microservice composition schemes according to the individualized fitness function, find the optimal microservice composition scheme among the n microservice composition schemes, and judge whether the corresponding fitness function value reaches the theoretical maximum excellent. If yes, then the microservice composition scheme is the global optimal microservice composition scheme; otherwise, execute S6. When the final optimal microservice composition scheme is found or the number of iterations reaches the upper limit, it stops, and the optimal microservice composition scheme at the time of stopping is output, which is the global optimal microservice composition scheme. Its specific description is as follows:

① When initializing n drive individuals, a list of drive individuals fireflies is created to store the drive individuals. By traversing the n micro-service combination schemes, the response function values of each micro-service combination scheme are calculated respectively, and the brightness attribute value of the light-driving individual is modified.

② Each light-driving individual needs to compare the brightness with other light-driving individuals. If there is a light-driving individual with higher light-brightness than this light-driving individual, the attraction degree between the two is calculated according to formula (11). Until the end of the comparison, according to the principle of maximum attractiveness, move to the light-driving individual that is brighter than yourself and has the greatest attractiveness.

③ Traverse the list of drive individuals, and assign the maximum fitness value to the global optimal fitness value fBest.

④ Determine whether fBest reaches the theoretical optimal value fTheory (this value represents the theoretical optimal moderate value of the service quality of the microservice combination under the corresponding target cost), if it reaches the theoretical optimal value, then X _index = (S ₀ ^(index) , S ₁ ^(index) ,…,S _m-1 ^(index) ) is the best microservice composition scheme. Otherwise, Count++ (the value of Count indicates the number of iterations, the initial value is 0), and judge whether Count reaches the maximum number of iterations MaxIteration, if so, output X _index to end the search; otherwise, execute S6.

S6: Update the micro-service combination scheme of 20% of the light-driving individuals with the chaotic disturbance rules, and these 20% of the light-driving individuals are to select 10% of the light-driving individuals with the best fitness value and 10% of the fitness value in the current iteration The worst optical drive individual, the rest of the optical drive individuals are still calculated according to the formula of the position transformation of the original combination algorithm. 20% of the CD-ROM individuals need to be updated, and then perform a local search. After the update is completed, return to S5. The specific steps are as follows:

①The 20% of the selected CD-ROM individuals need to update their positions according to the chaotic sequence. The formula is as follows:

In formula (19)

The update is obtained iteratively according to the cotangent sequence obtained at the tth time, where h represents the dimension of the individual driver, that is, the length of the service chain, and n represents the maximum number of iterations. The updated cotangent function value needs to be calculated according to S4 for the corresponding candidate service.

② 20% of the light-driving individuals conduct a local search to find their own optimal solution for the local microservice combination. Here, a chaos count, chaoNum, is set to limit the number of comparisons. When the light-driving individual finds the local optimum in the local search or exceeds this chaos number of times to end the search.

(3) The light-driving individuals participating in the chaotic disturbance need to compare the locally optimal combination scheme found by themselves with the historical global optimal combination scheme. If it is better than the historical global optimal combination scheme, update the historical optimal combination scheme with this scheme as the new fBest, otherwise continue to S5.

S7: The global optimal solution based on the current service chain is obtained through the combined algorithm, or the maximum number of iterations is reached, and the corresponding service chain is the current optimal service chain. After the microservice combination is successfully selected, the user's task is executed, and then the edge orchestration manager will return the execution result to the user, update the QoS attribute values in the corresponding microservice, and save the updated result in the Service Registry.

S8: The edge orchestration manager will save the QoS attribute values of the microservices registered in the service registry in the service registry, and use the hash tree concept to hash the QoS attribute values of the microservices in pairs, and finally get a hash root. The edge orchestration manager encrypts the hash value with the private key, and then broadcasts it to each edge server, and the edge server saves the hash value with the saved public key. The edge server will periodically ask the registry to see if the hash root stored in the registry is consistent with the locally stored hash value. If it is inconsistent, it is judged that the service QoS attribute value stored in the service registry has been tampered with, and the attribute value of which service has been modified can be quickly located by using the properties of the hash tree, and then the specific edge server can be located.

The above embodiments are only to illustrate the technical idea of the present invention, and cannot limit the protection scope of the present invention. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the present invention. Inside.

Claims

An edge cloud system-oriented software microservice combination optimization method, characterized in that it includes the following steps:

S1: The user submits his own task, which includes the corresponding task input and the expected output result. The task is submitted to the edge cloud, received by the edge orchestration manager, and deployed in each edge server through the registration center. Microservices information, and choose a combination of services;

S2: Introduce the predatory search strategy, and use the predatory search strategy to obtain a feasible service chain globally, that is, to search for a service chain that meets the requirements of the user's microservice composition;

S3: The length of the feasible service chain is m, and the candidate service corresponding to each microservice on the feasible service chain is searched, that is, the microservices with the same input set and output set as the microservices on the feasible service chain are put into the same service class , get m candidate service classes;

S4: In the candidate service class, use a cotangent sequence with chaotic properties to select a candidate service from each candidate service class in turn to form a micro-service combination. The micro-service combination is mapped to a driver individual, according to the previous driver The individual performs chaotic initialization n-1 times, which obtains n light-driving individuals;

S5: Evaluate the n microservice composition schemes according to the individualized fitness function, find the optimal microservice composition scheme among the n microservice composition schemes, and judge whether the corresponding fitness function value reaches the theoretical maximum If yes, then the microservice composition scheme is the global optimal microservice composition scheme; otherwise, execute S6, when the final optimal microservice composition scheme is found or the number of iterations reaches the upper limit, stop, and output the maximum value when stopped. Excellent microservice composition scheme;

S6: Update the micro-service combination scheme of 20% of the light-driving individuals with the chaotic disturbance rules, and these 20% of the light-driving individuals are to select 10% of the light-driving individuals with the best fitness value and 10% of the fitness value in the current iteration For the worst optical drive individuals, the rest of the optical drive individuals are still calculated according to the position transformation formula of the original combination algorithm. 20% of the optical drive individuals need to update their positions, and then perform a local search. After the update is completed, return to S5;

S7: The global optimal solution based on the current service chain is obtained through the combined algorithm, or the maximum number of iterations is reached, and the corresponding service chain is the current optimal service chain. After the microservice combination is successfully selected, the user's task is executed, and then the edge orchestration manager will return the execution result to the user, update the QoS attribute values in the corresponding microservice, and save the updated result in the Microservice registry;

S8: The edge orchestration manager will save the QoS attribute values of the microservices registered in the service registry in the service registry, and use the hash tree concept to hash the QoS attribute values of the microservices in pairs, similar to a tree structure Finally get a hash root. The edge orchestration manager encrypts the hash value with the private key, and then broadcasts it to each edge server. The edge server decrypts the broadcast information with the stored public key and saves the hash value. The edge server will periodically ask the registration center to see if the hash root stored in the registration center is consistent with the locally stored hash value. The attribute value of which service has been modified can be quickly located, and the specific edge server can also be located.
A kind of software microservice combination optimization method for edge cloud system according to claim 1, it is characterized in that: the concrete process of described S2 is as follows:

S2.1: Set the searchSet collection to be empty, search all microservices, add the microservices that meet the user's microservice composition requirements to the service chain, and store the input collection of the microservices in the searchSet collection;

S2.2: Before adding a microservice that meets the user's microservice composition requirements to the service chain, determine whether there is an input set of the microservice in the searchSet collection, if not, add the microservice to the service chain, otherwise, do not join ;

S2.3: After the current search is over, if no feasible service chain is obtained, the searchSet collection will be cleared, and S2.1-S2.2 will be repeated in the microservices that have not been searched until a feasible service chain is found or the maximum search is reached. times, stop searching.
The method for optimizing software micro-service composition for edge cloud systems according to claim 1, characterized in that: the specific process of S4 is as follows:

S4.1: The length of the feasible service chain is m, j=0,...,m-1, and the jth microservice has k j candidate services;

S4.2: Determine the order of magnitude of k j
According to the cotangent sequence value corresponding to the jth microservice
Truncate the cotangent sequence value with φ j bits after the decimal point as an integer value u i,j ,
n is the total number of drive units;

S4.3: Take the remainder of u i,j to k j to generate a chaotic value ξ i,j , the chaotic value corresponds to a candidate service in the candidate service class corresponding to the jth microservice, and this value corresponds to the If a certain service in the candidate services corresponds, the corresponding candidate service is selected. Similarly, after all services in the service chain perform this operation, the initialization of the position X i of the i-th optical drive individual is completed.
The software microservice combination optimization method oriented to an edge cloud system according to claim 1, wherein the calculation formula of the fitness function value in the S5 is:

Among them, f represents the fitness value, R represents the product of the reliability of each microservice on the service chain, T represents the sum of the response times of each microservice on the service chain, C represents the sum of the execution costs of each microservice on the service chain, and A represents the service The product of the availability of each microservice on the chain, α represents the weight of R, β represents the weight of A, γ represents the weight of T, η represents the weight of C, and α+β+γ+η=1.
The method for optimizing software micro-service composition for edge cloud systems according to claim 1, wherein the specific process of S6 is:

S6.1: The screened 20% of the light-driving individuals need to update their positions according to the chaotic sequence. The formula is as follows:

In the above formula
Represents the position of the driver i in the jth dimension of the t iterations, that is, the selection scheme of the jth component of the service chain. The position update also needs to be calculated according to the cotangent sequence. In the t+1th iteration, the ith The j-th dimensional position of each disc-drive individual
The update is obtained iteratively according to the cotangent sequence obtained at the tth time, where h represents the dimension of the individual driver, that is, the length of the service chain, and n represents the maximum number of iterations. The updated cotangent function value needs to calculate the corresponding candidate service according to S4;

S6.2: 20% of the CD-ROM individuals conduct a local search to find their own optimal solution for local micro-service combination. Here, a chaos count, chaoNum, is set to limit the number of comparisons. When the CD-ROM individuals find the local optimum in the local search Or if this number of chaos is exceeded, the search can be ended;

S6.3: Individuals participating in chaos disturbance need to compare the locally optimal combination scheme found by themselves with the historical global optimal combination scheme. If it is better than the historical global optimal combination scheme, use this scheme to update the history The optimal combination scheme is used as the new fBest, otherwise continue to S5.