- Service identification step is a basic requirement for a detailed design and implementation of services in a Service Oriented Architecture (SOA). Existing methods for service identification ignore the automation capability while providing human based prescriptive guidelines, which mostly are not applicable at enterprise scales. In this paper, we propose a top down approach to identify automatically services from business process. We use for clustering a hybrid particle swarm optimization algorithm and several design metrics for produce reusable services with proper granularity and acceptable level of cohesion and coupling. The experimental results show that our method HPSOSI (Hybrid Particle Swarm Algorithm for Service Identification) can achieve a high performance in terms of execution time, and significant quality in terms of high modularization, strong cohesion, and weak coupling of the identified services..
Frequent changes in the business environment
and user demands are two important challenges
in developing large-scale software systems.
Service-Oriented Architecture (SOA) is one of the
promising methods to address these challenges
[
The need for an automated approach to identify services is recognized to make service identification step more reliable.
Most of automated approaches use
Metaheuristic algorithms for service identification, such
as Simulated Annealing [6], Genetic Algorithm
[10], [
In this paper, a new identification approach is proposed, which aims to resolve the above problems by supporting automation capabilities, adopting technical metrics, and using business process as input. This method generates candidate software services using multi-objective hybrid particle swarm optimization algorithm in order to group them into distinct services represented as clusters by analyzing the dependencies between business activities and business entity. The proposed approach combines two Meta-heuristic algorithms that are PSO (particle swarm optimization) and GA (Genetic Algorithm). We use a hybrid algorithm by embedding the crossover and mutation operators of Genetic Algorithm into the PSO.
The rest of this paper is organized as follows. In section 2, the most related work is briefly reviewed. In Section 3 introduces to the basic concepts the particle swarm optimization. In section 4,a novel method is proposed in order to solve the service identification problem based on hybrid Particle Swarm Optimization. Section 5 provides an implementation and experimental results to demonstrate the performance of our approach. Finally, section 6 concludes the paper and point directions for future work.
The literature provides several works in service identification. Who uses various levels of automation: prescriptive, semi-automated, and fully-automated.
In this section, the existing approaches for service identification can broadly be separated into one of the following categories:
semi-automated):
Methods (prescriptive,
[5]mentioned some best practices, wherever
appropriate, to point out the vagueness involved
in service identification. A top-down and
bottomup techniques for service identification has
proposed in this methodology.
[9] presented a method of service identification
using an ontology for the product line. Primary,
Semantic relationship was derived through the
mapping between feature modeling and ontology.
Second, both service and service boundary was
defined by semantic distance. Third, the method
was proposed for feature grouping and candidate
service refining service candidate which is the
fittest service granularity.
[
In [10], an automated method for identifying business services by adopting design metrics based on top-down decomposition of processes is presented. This method takes a set of enterprise business processes as input and produces a set of non-dominated solutions representing appropriate business services using a multiobjective genetic algorithm. [6] introduced a novel approach called ASIM for automatically identifying and partly specifying enterprise-level software services for business models using best practices and principles of model-driven software development. They formulated service identification as a multiobjective optimization problem and solved it by a novel meta-heuristic optimization algorithm that derives appropriate service abstractions by using appropriate quantitative measures for granularity, coupling, cohesion, reusability, and maintainability.
In [
Particle Swarm Optimization (PSO), is a new
intelligent global optimization algorithm, it was
proposed in 1995 by Eberhart and Dr. Kennedy
[
PSO uses to solve an optimization problem, a
swarm of computational elements, called particles
to explore the solution space for an optimum
solution. [
PSO is suitable for continuous areas. [7], proposed HPSO (Hybrid Particle Swarm Algorithm) for solving problems in discrete areas by introducing the crossover strategy and the mutation strategy of the genetic algorithm. c r _ − x + c r _ − x is viewed as a crossover operator that make the current solution do crossover operation on individual extremum, and make the result do a crossover operation on global extremum. w × υ can be viewed as a mutation operator of genetic algorithm. The variation of the above results is a new location[7].
For our problem, we will use hybrid PSO by introducing the crossover and mutation operators as follow: (1) Crossover operator
We propose crossover operators that ensure consistency of solutions as follows:
1. Select a particle with a given crossover probability P .
2. For the current particle and a business process which is composed of n activities and m business entity, randomly select a position i, (where i = 1,…,n or i = 1,…,m).
3. The value of the selected position in the first sequence will exchange with the value of corresponding position in the second sequence. (2) Mutation operator
In order to guarantee the diversity of the population, we introduce the mutation strategy in genetic algorithm into PSO by randomly select x positions from the sequence, and replace it by a random number ranging in the interval [0...k], where k is the number of clusters.
Referring to the SOMA [1], different approaches can be adopted to identify services, namely top-down (domain decomposition), bottom-up (existing system analysis) and meetin-the-middle. The Top-Down approach consists mainly in decomposing business processes into finer business tasks. The bottom-up approach is about analyzing existing IT assets and finding functionality that could be exposed as services.
The last one is about to conduct both a bottom-up analysis and top-down analysis and to correlate the services identified by each of these approaches.
The goal of service identification from business process model is organizing a set of business activities into significant clusters (high cohesion and low coupled). The activities of a cluster are considered as the operations of a candidate service.
In our approaches, we begin with the business process model as input (Top-Down) and derive the effective service set based on this model by combining three techniques for grouping business activities in order to form software services:
putting all the activities associated with a particular business entity into a service. In this case, a CRUD (Create Read Update Delete) Matrix (CRUDM) is used as input in the clustering algorithm.
the activities performed by a particular actor into a service. In this case, an Activity-Actor Matrix (AAM) is used as input in the clustering algorithm.
putting all the activities those participate in the achievement of a particular business goal into a service. For this Technique, we use an Activity Goal Matrix(AGM) as input data in the clustering algorithm.
We use a directed graph to represent dependencies between business activities in order to make the structure of a complex business process more understandable.We call such graphs Activity Dependency Graph (ADG). In the ADG, the business functions (Business Activities, Sub Process) are represented as nodes and their relationships as edges that connect the nodes. Thus, an Activity A is connected with an Activity B if they manipulate the same Business Entity and/or are performed by the same Actor and/or they participate in the achievement of the same Business Goal. One way of making ADGs more available is to partitioned them by grouping closely related activities into clusters.
Definition 1 (Cluster) Let = ! , , , # $ , a set of Business Activities of a business information system. Each activity manipulates one or more Business Entity and is performed by one or more Actor and participate in the achievement of one or more Business Goal. Formally, a cluster Ci can be defined as: % = & ' | 1 ≤ j ≤| | where ∈ .
Definition 2 (Partition) A partition is a set of non-empty subsets of. Formally, a partition π can be defined as: π = {% } | 1 ≤ k ≤ | | Where ⋃ - % =
and ⋂ - % = ∅
In addition, a partition of into k non-empty
clusters is called a k-partition of . Given a set
that contains n elements, the number
0#, called Stirling numbers of distinct k-partition
of satisfies the recurrence equation presented
by [
An effective way of increasing the quality of
software products consists of using metrics to
guide the development process. For evaluating
and measuring the quality of the services
produced by our clustering technique in each
iteration, we use three design metrics introduced
by [
Definition 3 (Coupling between two cluster ) The Coupling Between two Clusters (CoupBC) measures the degree of connectivity between two distinct clusters. A high degree of interconnectivity is undesirable because it indicate that services are highly dependent on each other. Formally, %5> ?%@ (coupling between cluster a and cluster b) measures the ratio of interdependencies between cluster a and cluster b and the maximum possible number of interdependencies between those clusters.
0 23 C = D %5> ?%@ = A2 ∗ ∑P-L ∑P-N GHIJKLM, JKNO 23 C ≠ D!2$ Where Q@is the number of activities of cluster a, Q is the number of activities of cluster b, and the value of GHIJKLM, JKNO (Total Dependency Matrix) represents the degree of dependency between the 2S activity of cluster a ( % @M ), and the TS activity of cluster b ( ACWX ).
Definition 4 (Coupling Of a given Cluster) The Coupling Of a given Cluster (CoupOC) measures the degree of connectivity between a given cluster and all other clusters of the system. Formally, a Coupling Of a given Cluster a is defined as:
0 %5> Y%@ = Z ∑N`bac K\]^_K LN
Pa7
Service cohesion refers to the degree of the
strength of functional relevance of activities
carried out by a service to realise a business
process [
Formally, a Cohesion Of a Cluster a (%5ℎY%@) is defined as: %5ℎY%@
0 = d∑#-7 ∑#0 Where %5ℎY%@ is the cohesion of the cluster a, and %5> Y%@ is the coupling of the cluster a. Definition 7 (Modularization Quality) Modularization Quality (MQ) determines the quality of an ADG partition perfectly as the tradeoff between interconnectivity (i.e., dependencies between the activities of two distinct services) and intra-connectivity (i.e., dependencies between the activities of the same service). This trade-off is based on the hypothesis that well-designed service-oriented software systems are organized into cohesive subsystems that are loosely interconnected. Therefore, a high MQ allows the creation of highly cohesive clusters that are not coupled strongly.
Formally, The Modularization Quality (MQ) is defined as the average of the MF’s of all clusters: # 1 Ik = ∗ l Ig 6 Where n is the total number of cluster in the system. And Ig is the Modularization Factor of the 2S cluster.
The service identification process starts by querying the Business Process Ontology to extract the existing dependencies between Activity, Business Entity, Actor, and Goal concepts. Three matrices are created to represent the extracted dependencies: (i) the CRUD Matrix (CRUDM) which describes the relations between the Activities and the Business Entities, (ii) the AAM (Activity-Actor Matrix) which describes the relations between the Activities and the Actors, (iii) and the AGM (Activity-Goal Matrix) which describes the relations between the activities and Business Goals (cf. Figure 1).
4.2.1
Activity Relationships Matrices The CRUD Matrix. For grouping Business Activities into clusters using the Business Entity Centric Technique, we need a rate that represent the degree of semantic relationships between Activity and Business Entity. Four operations can be applied on each Business Entity by the Business Activity. These operations are called CRUD (C:Create, R:Read, U:Update, D:Delete). In order to compute the value of each technical metrics, such as Cohesion, Coupling, and Modularization Quality, it is necessary to associate each operation with a value between 0 and 1. The majority of existing works on CRUD semantic relationships adopts the following substitutions: C=1; U=0.75, D=0.50, R=0.25. These values represent the cost of the operation on each Business Entity.
We store these values in a CRUD Matrix (CRUDM), which has Business Activities as its rows, Business Entities as its columns, and semantic relationships (”C”, ”U”, ”D”, ”R”, with the distinct intensity 1 => C > U >D > R > 0) as its cells. Each column in this matrix must have exactly one create operation and each row must have no conceptual (operational) activities. Table 1 shows an example of CRUDM.
Activity-Actor Matrix. In order to regroup activities using Actor-Centric Technique, we represent the degree of semantic relationships between Activity and Actor by using a second matrix called ActivityActor Matrix (AAM). The value of each cell of AAM is taken between 0 and 1. For example the value AAM1,3 = 0.75 indicates that the Actor 3 perform 0.75% of the Activity1. The sum of each row of AAM must be equal 1 to indicate that the corresponding activity is performed completely. Activity-Goal Matrix. In order to group Business Activities based on Goal-centric Technique, we use a third matrix called Activity-Goal Matrix (AGM). Each cell of AGM represents the degree of semantic relationships between Activity and Business Goal. The value of each cell of AGM is taken between 0 and 1. It indicates the achievement rate of a Business Goal after the execution of a Business Activity. For instance the value AGM2,4 = 0.50 indicates that the activity 2 Achieve 50% of the Goal 4. The sum of each column of the AGM must be equal 1 to indicate that the corresponding Business Goal is achieved completely.
derivation In order to enhance the quality of service cluster, three square matrices are derived from CRUDM, AAM, and AGM respectively. The first one is called Business Entity centric Activity Dependency Matrix (BEADM), the second one is called Actor-centric Activity Dependency Matrix (AADM), and the third one is called Business Goal centric Activity Dependency Matrix (BGADM). These square matrices represent the degree of dependency between each activity and other activities of the annotated business process. To compute dependencies between two activities, we take the minimal value of relationships that relies the both activities with the same Business Entity.
For instance, the degree of dependency between Activity 1 that read Business Entity 5 and Activity 3 that create the same Business Entity 5 is equal 0.25 (we use the minimal value between R:0.25 and C:1). If no operation is applied by the Activity 2 on Business Entities that are manipulated by Activity 1, we conclude that there is zero dependency between Activity 1 and Activity 2. The three matrices BEADM, AADM, and BGADM are integrated into one matrix named Total Dependency Matrix (TDM) that represents the average dependency between each couple of activities. Each cell of TDM is calculated as follows: = TDM ,q
W ∗ BEADM ,q + W ∗ AADM ,q + Wu ∗ BGADM ,q ∑u- W Where W1,W2, and W3 are weighting factors corresponding to BEADM, AADM, BGADM respectively.
4.2.3 Hybrid Particle Swarm OptimizationBased Clustering.
Matrix clustering is considered as NP-complete problem, we have to use Meta-heuristics to solving it.
We propose Hybrid Particle Swarm Optimization [7] for clustering the TDM matrix.
The steps of HPSOSI(Hybrid Particle Swarm Optimization for Service Identification)is mainly divided into three parts: • First, initialize the particles ; • Second, calculate the fitness value for each particle and update pbest and gbest; • Third, use the crossover and mutation operation to improve PSO.
The specific steps are shown in Figure 2. # 1 3296;== = Ik ∗ w∗ l %5ℎY% x ∗ !1 − ! 6 #∗ l %5> Y% $$
Where n is the total number of clusters represented by the current particle, %5ℎY% is the cohesion of the the ith cluster, and %5> Y% is the coupling of the jth cluster. A better particle is a particle that has a strong quality of modularization, a strong cohesion, and a weak coupling which implies a high fitness.
In order to evaluate the performance of our approach we implemented a tool called HPSOSI, Figure 3 is shown a printout screenshot of this latter, which was developed in order using Java programming language with NetBeans IDE 7.0. All experiments run in Windows 7 on a desktop PC with Intel Dual Core, 2.3 GHz processors, 2GB of RAM.
The HPSOSI parameters were set experimentally as follows: crossover probability Pc and mutation probability Pm we reset to 0.8 and 0.4 respectively and weighting factors W1=1,W2=2,and W3=1. The swarm comprises of 1000 particles and the number of iteration k = 100.
To evaluate the performance of our tool HPSOSI (Hybrid Particle Swarm Optimization for Service Identification), we applied the proposed clustering algorithm on three business processes with different combination of weighting factors. The provided results are shown in Figure 4.
We note that some factor influences more than others on the fitness value, because the summation of total semantic relationships for matrices CRUDM, AAM, and AGM are different. There is especially for total semantic relations (TSR) in the matrix AAM is still a higher summation of total semantic relations for CRUDM or AGM matrices. They can be formulated as follows:
G0y! I$ > G0y !%yzHI$ > G0y ! {I$.
G0y! I$ = #€_
#_ l 0y!}?~ $ G0y!CRUDM$ = l 0y!?~ $
#…†J‡ G0y!AGM$ = l 0y!{Y „ $
0y!}?~ $ = 0y!?~ $ = 0y!{Y „ $ = 1. 0y!}?~ $ : Summation of semantic relationships in EBPi row for I . 0y!?~ $ : Summation of semantic relationships in BPi Column for %yzHI. 0y!{Y „ $ : Summation of semantic relationships in GOALi Column for {I .
The business process of the case study is
shown in Figure 5.
We run our tool on the same business process
used in [6], [8] and [
Although we have the same number of iterations
in [
Matrix clustering for services identification is an NP-complete problem. This paper presents HPSOSI a multi-objective hybrid particle swarm optimization algorithm, where the crossover and mutation of a genetic algorithm is embedded into the PSO to improve the local search ability and to maintain the diversity of the population in order to achieve better performance than standard evolutionary algorithm. The proposed method identifies automatically candidate SOA services from business process models.
The experimental results show that our approach achieves high performance in term of convergence speed and execution time compared with other solution.
As future work, we are studying introducing others semantic annotation to fully consider semantic relationships between business elements, and hence transaction and semantic integrity can be guaranteed.
[1] Arsanjani. Service-Oriented Modeling and Architecture (SOMA). IBM developer Works, 2004. [2] Azevedo Leonardo Guerreiro, Flávia Santoro, Fernanda Baião, Jairo Souza, Kate Revoredo, Vinícios Pereira, and IsoldaHerlain. “A Method for Service Identification from Business Process Models in a SOA Approach“. CAiSE, pp. 99-112, 2009. [3] Alejandro Cervantes. Inés Galván. Pedro lsasi.Binary particle swarm optimization in classification. Neural Network World 3(05), pp. 229-241,2005. [4] Erl T. Service-Oriented Architecture: Concepts, Technology, and Design. Prentice Hall PTR, 2005, 792 p. [5] Inaganti S. and Behara G. K .Service Identification: BPM and SOA Handshake.
White Paper,2007. [6] Jamshidi P. Mansour S. Sedighiani K.
Jamshidi S. and Shams. “An Automated Service Identification Method“.Technical Report. Shahid Beheshti University.2012. [7] Sheng-Jun Xue, Wu Wu. Scheduling Workflow in Cloud Computing Based on Hybrid Particle Swarm Algorithm. Nanjing University of Information Science & Technology, School of Computer and Software, Nanjing, China.TELKOMNIKA, Vol.10, No.7, November 2012, pp. 1560~1566. [8] Chergui Mohamed El Amine, Sidi Mohamed Benslimane. Using Combinatorial Particle Swarm Optimization to Automatic Service Identification. 13th International Arab Conference on Information Technology ACIT’2013, December 17-19, 2013, Khartoum, Sudan. [9] Kang DongSu, Chee-yang Song, Doo-Kwon Baik. “A Method of Service Identification for Product Line“. Proc. Of Third 2008 International Conference on Convergence and Hybrid Information Technology, pp. pp. 1040–1045, 2008. [10] Kazemi Ali, Ali Rostampour, PooyanJamshidi, Eslam Nazemi, Fereidoon Shams, Ali Nasirzadeh Azizkandi. “A Genetic Algorithm Based Approach to Service Identification“.IEEE World Congress on Service Computing, Washington DC, USA, July 4-9, 2011
Conference