CN109408046B

CN109408046B - Shortest-path Web service combination method based on graph

Info

Publication number: CN109408046B
Application number: CN201811031345.8A
Authority: CN
Inventors: 许国艳; 李政道; 程义; 朱进
Original assignee: Hohai University HHU
Current assignee: Hohai University HHU
Priority date: 2018-09-05
Filing date: 2018-09-05
Publication date: 2022-01-28
Anticipated expiration: 2038-09-05
Also published as: CN109408046A

Abstract

The invention discloses a shortest-path Web service combination method based on a graph, which changes a KBG (KBG) search algorithm into a heuristic A-algorithm aiming at the problems of overlarge time and space cost and excessive redundant service based on the KBG algorithm, can accelerate the search speed by selecting an heuristic function, simultaneously deletes and merges equivalent services, deletes and merges service nodes by the inclusion relationship of service parameters, and reduces the number of the nodes and deletes redundant nodes. The complexity of generating the graph is reduced by detecting and combining equivalent services and the number of compression nodes, the efficiency of the algorithm is improved, and the method has lower space resource consumption and time consumption.

Description

Shortest-path Web service combination method based on graph

Technical Field

The invention relates to a shortest Web service combination method based on a graph, belonging to the technical field of Web service combination in the field of Web services.

Background

The Web service is an independent software component which can complete the designated function, and can carry out data exchange between different programs across platforms. The description language employed by the Web service is an XML standard language, which allows it to be described, published, discovered, and used. The Web service can be described as a software functional module which is independent of a platform, low in coupling, secondary in development and self-contained in nature, and each Web service can independently complete a specific functional task. The relevant standard specification of the Web service tends to be mature after continuous development and perfection, and is gradually accepted and adopted by enterprises and applied to production. However, in many cases, a single Web service may have difficulty meeting complex requirements of users, and in this case, if the results required by users can be obtained by screening and combining existing services, the results are far better than those obtained by newly designing a Web service meeting the requirements of users. The Web service combination brings some advantages, a plurality of atomic services are combined into a service with larger granularity, the service with larger granularity can be reused by other services, and the combination system can also be used as other service combination systems for combination, so that the Web service combination has multilayer and expansibility. In addition, the method is dynamic and self-adaptive, can be modified according to the requirements of users before combination or in the combination process, and can be self-adaptive without influencing the operation of the next step. At present, some service combination methods, such as workflow-based combination, semantic network-based combination and other methods, have been proposed, but because of some disadvantages that the methods are not very adaptive, relatively complex and not easy to implement, and the like, there is an urgent need for a Web service combination method that is easy to implement and has less service preprocessing.

At present, the research field of Web service composition is mainly based on a graph method, and although the composition schemes are different, the main ideas of the methods are consistent. In the graph-based service composition method, a single Web service is regarded as a node in the graph, and the input and output of the node can correspond to the input and output parameters of the Web service. A service dependency graph is typically constructed using the available services and a service composition path corresponding to the user requirements is searched. In the KBG algorithm mentioned in Kona, a service dependency graph is first generated from a service library. The graph is calculated iteratively, all possible services which can be called in the service library are added into the current layer in each step, then the dependency graph is cut according to the input and output requests of the user, and finally construction and search are carried out. Although useless services are filtered, the algorithm does not guarantee that the optimal combination can be found, and meanwhile, the time cost and the space cost are high, and a certain optimization space is provided.

Disclosure of Invention

The purpose of the invention is as follows: aiming at the problems in the prior art, the invention provides an A-algorithm-based Web service combination method (WSCBA) with lower space resource consumption and time consumption by combining a Web service combination technology and a KBG algorithm.

The technical scheme is as follows: a shortest Web service combination method based on a graph introduces an A-algorithm into a Dijkstra algorithm with low efficiency in a KBG algorithm, adds a redundant service processing technology which is not related to the KBG algorithm, and adopts a technology of compressing the number of nodes in the execution process of the A-algorithm, and specifically comprises the following steps:

step 1: generating a Web service dependency graph D according to the service information list, and adding the Web service dependency graph D into the virtual node dummy service; secondly, performing BFS once on the Web service dependency graph added into the virtual node from the positive direction according to the Web service input parameter Sin and the Web service output parameter Sout, deleting part of useless services, performing BFS once again from the negative direction, and finally generating a dependency sub-graph H;

step 2: detecting and merging equivalent services according to the Web service dependent subgraph H, and reducing the number of nodes;

and step 3: and performing node compression operation on the processed H, introducing an A-x algorithm to search a path, and detecting and compressing neighbor nodes of the composite nodes in the process.

In the step 1, besides the need of constructing the Web service dependency graph, a method for constructing a Web service dependency subgraph is provided on the basis of the Web service dependency graph, and all paths meeting the given service query are extracted. The specific process is as follows:

(1) the original input of all the services is used as an input set, the services which are set as input parameter sets into proper subsets of the input set are added into the first layer by circularly traversing the service library, and the output parameters of the newly added services are added into the input set after circulation is finished, so that the steps are repeated;

(2) if a service S ∈ L_iDoes not receive its nearest previous layer L_i-1The generated input o, then we are at L_i-1A virtual node D is inserted, the virtual node D takes o as input and generates the same output, and a virtual service (namely the virtual node D) connects S with the service which originally generates o to obtain a Web service dependency graph D;

(3) according to the Web service dependency graph D obtained in the step (2) and the user request, deleting the services which do not meet the input conditions by using a BFS algorithm (breadth-first traversal algorithm) with the user input parameters as targets in the positive direction;

(4) and deleting the services which do not meet the output condition by using a BFS algorithm with the output parameters as targets from the reverse direction to obtain a Web service dependent subgraph H.

In the step 2, the processing of the redundant service and the equivalent service in the generation path is added, a redundant service processing technology is provided, the equivalent service is combined, and the number of nodes is reduced. The specific process for detecting the merged equivalent service comprises the following steps:

(1) for each layer of the Web service dependent subgraph H, putting the services providing the same input parameters into a group;

(2) firstly, comparing QoS values of services in a group;

(3) for the service group Si to be superior to Sj, let listi and listj be the sets of services with Si and Sj as predecessor nodes, respectively. If the listi contains listj, going to the next step, otherwise Si and Sj can not be combined; and comparing the Qos values of the service groups Si and Sj, and judging which is more optimal according to the Qos values, wherein the Qos values refer to response time (responseTime), the response time is used as a basis for judging the optimal, and the service group has less response time and is more optimal.

(4) And comparing whether Siout contains Sjout, if so, deleting Sj, otherwise, Si and Sj can not be combined, and the Siout and the Sjout are respectively the output of the service groups Si and Sj.

In the step 3, an a-algorithm is introduced, the search algorithm of the KBG is changed into a heuristic a-algorithm, the search speed can be increased by selecting an heuristic function, and nodes are compressed at the same time.

A, algorithm construction: n is composed of a group of nodes (N)_i) Formed composite service, wherein N_iIs a layer L_iA set of services in (1). One goal of Web service composition is to minimize the number of atomic services of a composite service, where the cost function of a service composition is defined as the number of services, denoted g (n):

L_Nis the first layer of the current combinational service, # L is the first layer of the dependency graph, and the cost function represents its slave node group N_iThe number of services is counted, but the virtual services in the node are not counted.

The heuristic function should estimate the cost of the solution. The layer number represents the distance from the initial node using the layer where the node is located as a heuristic. The service in layer 3 means that the algorithm requires more than three steps to reach the initiating node. The heuristic function is defined as:

h(N)＝distance(N_i) (2)

distance(N_i) Representing an optimal path to a target nodeThe estimated cost of the path is calculated by using a Hamamandon algorithm, and the estimation function f (N) of N can be defined as follows:

f(N)＝g(N)+h(N) (3)

the algorithm a has been introduced into a new Web service dependency sub-graph H, and during the execution of the algorithm a, the current node and its neighboring nodes are detected and merged if there is an equivalent node. The equivalence node needs to satisfy two conditions: firstly, the f (N) values of the two nodes are the same; second, the predecessor service graphs for both nodes are the same.

By adopting the technical scheme, the invention has the following beneficial effects: firstly, the Web service dependency graph is constructed according to the service library, the obtained service dependency graph is stored, when the service library is kept unchanged, when a service combination request is initiated, the operation is only needed to be carried out on the graph according to the requirements of users, and secondary generation is not needed. Secondly, the Web service dependency subgraph is obtained by cutting on the basis of the dependency graph D, and the main purpose is to narrow the search range and remove useless services. In addition, the detection and combination of the equivalent service algorithm also aims to reduce the search space, and the node compression technology also aims to reduce the complexity of the graph so as to improve the search speed.

Drawings

FIG. 1 is a flow chart of the WSCBA method;

FIG. 2 is a schematic diagram of Web service dependencies;

FIG. 3 is a graph of service number versus experimental results;

FIG. 4 is a graph of the results of the corresponding time-contrast experiment;

FIG. 5 is a graph showing the results of the combination elapsed time comparison experiment.

Detailed Description

The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.

The implementation of the invention adopts java programming language and eclipse mars platform, wherein the specific PC configuration is CPU-i 52500, DDR3-8G, SATA6-1TB, operating System-Win 7. The method was evaluated against 8 public repositories proposed by Web service challenge (wsc-2008) in 2008.

The experimental thought is as follows:

in the experiment, the service description file service.wsdl and the service quality description file service level elements.wsla provided by the data set are analyzed to obtain service description information and service QoS information. Wsdl file is provided by the service set, and is a designed request file containing specific service request parameters. Meanwhile, the best combination index values can be obtained at present, and the best values are added into the experimental result. And simulating a real environment, analyzing the service information, combining the services according to the requirements of the request file and recording various data indexes.

The shortest Web service combination method based on the graph comprises the following steps: firstly, traversing a service library to generate a service dependency graph D, and adding virtual nodes; secondly, removing useless services by using a BFS algorithm twice according to requirements, and generating a dependent subgraph H; then, detecting and merging equivalent services on subgraph H; finally, a path is searched on the processed H using the a-algorithm and the number of nodes is compressed.

The original KBG algorithm omits the processing of redundant services and equivalent services, the generated paths may contain the redundant services, and redundant nodes can expand the dependency subgraph in an exponential level. Aiming at the problem that the KBG algorithm does not effectively control the number of generated graph nodes, an improved Web service combination method (WCSBA) based on the A-x algorithm is provided. The method is concretely realized as follows:

1. constructing a service dependency graph

The service dependency graph shows the input and output dependency relationship among the services, and each node contains one service. If N is present₁The output of the service in (1) is at N₂A slave node N may be added at the time of input of the service(s) in (1)₁To node N₂Has a directed edge E. The first layer (L1) of the dependency graph consists of a set of nodes containing services, the inputs of which belong to a set of native inputs. Such asFIG. 2 shows a service dependency graph, the second level (L2) of which consists of a set of nodes containing services whose inputs belong to the native inputs of the first level and the outputs they produce. The layers of the dependency graph are mutually exclusive, i.e. no two layers of dependency graph contain a common service, which avoids the occurrence of loops in the graph, i.e. the case of infinite loops.

The Web service dependency graph is generated by a step-by-step iteration in layers. In the first layer, the native inputs of all the services are used as an input set, the services with the input parameter set being a proper subset of the input set are added into the first layer by circularly traversing the service library, and the output parameters of the newly added services are added into the input set after the circulation is finished, so that the steps are repeated; and finally adding virtual nodes into the graph. The Web service dependency graph generation algorithm pseudo-code is as follows:

2. structural dependency subgraph

However, not all paths in the dependency graph satisfy a given query, so this experiment extracts a subgraph of the service composition dependency graph, which contains all paths satisfying the given service query. For the structure of a subgraph H of a dependency graph D for querying R < Ri, Ro >, finding the subgraph H twice on the dependency graph D by using a breadth-first search (BFS) algorithm, the specific steps are as follows: in a first step, BFS is applied to the forward direction, and services that do not meet the input conditions are deleted from the start of inputting services belonging to Ri; next, BFS is applied again from the services whose output belongs to Ro, and services that do not meet the output conditions are deleted. The Web service dependent subgraph generation algorithm pseudo-code is as follows:

3. detecting and merging equivalent services

The detection and combination of the equivalent service algorithm is also used for reducing the search space, in the experiment, a subgraph H is depended on for the service, and the number of nodes is reduced by combining the equivalent service through detecting the equivalent service. For each layer of H, services that provide exactly the same input parameters are put into a group. For each group, we do the following: firstly, comparing QoS values of services in a group; for a service group Si superior to Sj, leading listi and listj to be respectively a set formed by services taking Si and Sj as precursor nodes, if the listi contains the listj, proceeding to the next step, otherwise, Si and Sj can not be combined; and comparing whether Siout contains Sjout, if so, deleting Sj, otherwise, Si and Sj can not be combined. The detection of the merged equivalent service algorithm pseudo-code is as follows:

4. compression node

In order to reduce the complexity of the graph to increase the search speed, we need to perform a compression node operation. The a-algorithm was introduced in the experiments because of the insufficiently efficient Dijkstra algorithm in the KBG algorithm. In this experiment, the compression node mainly performs the following operations: creating a new graph G; adding a virtual node as the output required by the user; searching each input parameter of the node N for a service capable of providing the parameter, and putting the service into a corresponding list; combining the services of all lists, finding out all possible combinations, and establishing a node N for each combination_i(ii) a A, executing an algorithm, and compressing equivalent nodes; adding the rest nodes into the L-1 layer of the graph G; repeating the above operations until the node containing the user input parameters is searched, wherein G at the moment is the graph formed by the service path. The compression node algorithm pseudo code is as follows:

the following examples were analyzed with reference to the accompanying drawings:

the experimental part mainly analyzes the improvement effect of the algorithm from three comparative experiments, and the comparative factors are respectively as follows: the number of services participating in the combination, the combined service response time, and the combined elapsed time.

For the user, the fewer atomic services involved in the composite service, the better. Invoking fewer services means higher reliability and there is a high probability that redundant services exist in the composite service. Therefore, the number of participating combination services can measure the accuracy and the redundant processing capability of the combination algorithm. In fig. 3, the operating result of the KBG algorithm and the operating result of the WSCBA method are respectively shown, and the best combination situation in the service set is also shown. It can be seen from fig. 3 that the WSCBA approach achieves the best combination of the data sets, except for the sixth and eighth data sets, and that there are almost redundant services in the KBG algorithm. Therefore, in this experiment, the WSCBA method is stronger than the KBG algorithm in terms of redundant service handling.

The response time of the combined service can better depict the construction precision of the Web service combination algorithm by taking the response time as the only QoS factor to participate in service combination. As can be seen from fig. 4, the difference between the WSCBA method and the KGB algorithm is small at a small data scale, and a certain difference begins to appear in response time at a large data scale. The WSCBA method does not perform as well as the KBG algorithm in the fourth and sixth data sets because the WSCBA method, although considering QoS optimality, mainly acts on the a x algorithm search process, mainly considering the minimization of the number of combinational services, and thus is slightly insufficient in QoS optimality. The WSCBA method is superior to the KGB algorithm in performance on the seventh data set, and the effect on the QoS constraint part in the WSCBA method is also reflected. In summary, the WSCBA method does not perform as well as the KBG algorithm in response time, but there is room for improvement.

In the case of guaranteeing the quality of service composition, it is more desirable that the algorithm has a faster execution time. Because algorithm execution time has a great relationship with the operation environment, the optimal time comparison is not added in the experiment, and only the execution conditions of the WSCBA algorithm and the KBG algorithm in the local are compared. As shown in fig. 5, the third, fifth, seventh and eighth data sets KBG algorithm show large fluctuation, which indicates that the KGB algorithm becomes less efficient to execute in the case of large data volume; the WSCBA method appears to fluctuate more smoothly under the same data set. The WSCBA method performed better than the KBG algorithm on all datasets in terms of combinatorial elapsed time. Therefore, the WSCBA algorithm performs more efficiently than the KBG algorithm under the present data set.

The performance analysis shows that the method has good effect, and the WSCBA method has lower space resource consumption and time consumption than the KBG algorithm.

Claims

1. A shortest Web service combination method based on a graph is characterized by comprising the following steps:

step 1: generating a Web service dependency graph D according to the service information list, and adding the Web service dependency graph D into a virtual node; secondly, performing BFS once on the Web service dependency graph added into the virtual node from the positive direction according to the Web service input parameter Sin and the Web service output parameter Sout, deleting part of useless services, performing BFS once again from the negative direction, and finally generating a dependency sub-graph H;

step 2: detecting and combining equivalent services according to the Web service dependency subgraph H; the processing of redundant services and equivalent services in the generation path is added, a redundant service processing technology is provided, the equivalent services are combined, and the number of nodes is reduced;

the specific process for detecting the merged equivalent service comprises the following steps:

(2) firstly, comparing QoS values of services in a group;

(3) for the service group Si superior to Sj, let listi and listj be the set formed by the services with Si and Sj as precursor nodes respectively; if the listi contains listj, going to the next step, otherwise Si and Sj can not be combined; judging who is better according to the Qos values by comparing the Qos values of the service groups Si and Sj;

(4) comparing whether Siout contains Sjout, if so, deleting Sj, otherwise, Si and Sj can not be combined, and Siout and Sjout are respectively the output of the service groups Si and Sj;

and step 3: performing node compression operation on the processed H, introducing an A-algorithm search path, and detecting and compressing neighbor nodes of the composite node, wherein the specific steps are as follows: changing the search algorithm of the KBG into a heuristic A algorithm, accelerating the search speed by selecting an heuristic function, and compressing nodes at the same time;

a, algorithm construction: n is composed of a group of nodes (N)_i) Formed composite service, wherein N_iIs a layer L_iA set of services in; one goal of Web service composition is to minimize the number of atomic services of a composite service, where the cost function of a service composition is defined as the number of services, denoted g (n):

2. The graph-based shortest Web service composition method of claim 1, wherein in step 1, a method for constructing a Web service dependency subgraph is proposed on the basis of a Web service dependency graph, and all paths satisfying a given service query are extracted; the specific process is as follows:

(2) if a service S ∈ L_iDoes not receive its nearest previous layer L_i-1The generated input o, then we are at L_i-1A virtual node D is inserted into the system, the virtual node D takes o as input and generates the same output, and the virtual service connects S with the service which originally generates o to obtain a Web service dependency graph D;

(3) according to the Web service dependency graph D obtained in the step (2) and the user request, deleting the services which do not accord with the input conditions from the positive direction by using the BFS algorithm with the user input parameters as targets;

3. The graph-based shortest Web service composition method of claim 1, wherein the heuristic function is defined as:

h(N)＝distance(N_i) (2)

distance(N_i) The estimated cost, representing the best path to the target node, is calculated using the hamaton algorithm, as defined above for the evaluation function f (N) of N:

f(N)＝g(N)+h(N) (3)

the A-algorithm is introduced into a new Web service dependency sub-graph H, and in the A-algorithm execution process, a current node and neighbor nodes thereof are detected and merged if equivalent nodes exist; the equivalence node needs to satisfy two conditions: firstly, the f (N) values of the two nodes are the same; second, the predecessor service graphs for both nodes are the same.