The Business process achievement requires various and heterogeneous knowledge i.e. computer information, business and information technology. Enterprises relying on Business processes to gain an added value should have a suficient expertise necessary to maintain their competitiveness in the marketplace. The cloud computing and more specifically the Software as a Service (SaaS) composition, is a prominent solution allowing the well achievement of Business Processes (BPs) goals and objectives. However,composing multiple SaaS services to support the BP execution is not a trivial task regarding the complex nature of the BP. Indeed, BP is composed of a set of inter-related activities having each one a specific functional requirements, data that should be exchanged between each SaaS service, and roles imposing that only authorized actors can perform a specific activity. This paper proposes a comprehensive framework for SaaS selection and composition in order to cover various considerations related to matching BP activities requirements and SaaS functionalities. The allocation of BP activities to be executed SaaS services using a genetic algorithm while considering the data and the activities access control issues.
Achieving " the enterprise business goals and their strategic objectives is becoming easier regarding the maturity of the available Business Process Management Systems (BPMS). However, deciding to cut of with traditional Information Technology (IT) systems to adopt BPMS ones is not a trivial neither a well-worn practice. The reconstitution of the enterprise IT to adopt business process based systems requires good business and IT expertise.
Taking advantage of the maturity of emergent and available technologies, such as cloud computing services, is a less expensive solution, as it reduces the burden of installing and configuring BPMS. Integrating cloud computing in enterprise IT assets is exhibiting exponential interest. Indeed, cloud computing is seen as a potential means to increase enterprises’ business agility, reduce software and hardware expenditure, and enhance productivity.
Regarding the plethora and the proliferation of the Software as a Service (SaaS) in the cloud environment, finding the most appropriate service, to support the achievement of the enterprise functional and non functional requirements, is becoming a fastidious task. For enterprises relying on business process to gain added value, considering several SaaS services cooperating to achieve the added value, requires tackling several issues. Firstly, each SaaS ofers diferent business functions which may fit or not each Business Process (BP) activity functionality. Moreover, there is no standard SaaS services description that unifies the heterogeneity when labeling the SaaS oferings. Furthermore, the business process structure, makes harder the BP outsourcing to SaaS. Indeed, BP is composed among others, of flow and connecting objects, swimlane and artifacts. Adopting BP based SaaS composition should keep as most as possible the BP main components.
This paper proposes a comprehensive system for BP based SaaS composition that tackles the overall mentioned issues : • The semantic matching between each BP activity and its data object with the SaaS ofering and its input/output data. Indeed, a BP activity/data label and the SaaS service/data may use diferent semantic description although they are functionally equivalent. • The consideration of the data exchanged between the on-premise BP activities and the outsourced to SaaS ones. Indeed, the access to data managed by SaaS ofers is not always possible. Furthermore, importing and exporting data to and from SaaS services may be not allowed for some SaaS providers. Hence, accessing to data handled by SaaS providers is very critical in the BP based SaaS composition as diferent SaaS services are supposed to cooperate together and/or with on-premise activities to achieve the BP functional objectives. • The consideration of role based activities. Indeed, the role concept is very important in the BP context. Each activity is performed by an actor to whom is assigned a role. Hence, controlling who is authorized to perform the BP activities in the SaaS environment is of a prominent importance. This implies assessing whether the SaaS service supporting the BP execution provides role based access control techniques.
Selecting the most appropriate SaaS service to support business process execution, is a large
scale optimization problem that can be resolved by applying meta-heuristic methods. Hence,
we propose the use of the Genetic Algorithm (GA) [
The sequel of the paper is as follows : Section 2 details our proposed system along with its diferent modules, Section3 gives an overview of the related work, Section 4 evaluates our system. In order to subjectively evaluate this research work, we present the threat to validity in Section 5. We conclude finally in Section 6.
The most prominent gathered data related to SaaS services are: the service provider, the ofered functionalities, the data access and the role based access control techniques.
The proposed system follows four main activities, namely : The BP re-construction, the semantic matching computing , the SaaS composition and finally, the SaaS modeling activities.
To guaranty keeping as much as possible the business process control flow and to simplify
the outsourcing task, the BP re-construction aim is to decompose the business process into a
hierarchy of sub-process having each one a Single Entry and a Single Exit (SESE) [
As the Figure illustrates, the BP encompasses five SESE fragments. The fragment 1 is the root of the sub fragments 2 and 5. Hence, we impose as constraints that if the root is outsourced, its sub-fragments cannot be outsourced and if the sub-fragment is outsourced, the root cannot be outsourced. This constraint is imposed to guaranty a correct outsourcing without overlaps. This constraint is considered in the cloud services composition activity to guarantee a well synchronized BP execution while supported by cloud services.
We propose in this section, a semantic process aiming to match each BP activity with its
data object with the SaaS ofering and its input/output data. More specifically, we propose a
process based on the extraction of semantics from the SaaS service Web pages to match the
appropriate SaaS service to each BP activity. To do so, BP activities and SaaS services should
have similar syntactic patterns. To attain this objective, we concentrate on a combination
of statistical, linguistic as well as semantic analysis. Regarding that each SaaS Web pages
contains a set of services which is by itself composed of a set of sub-services, it is essential to
depict some hidden knowledge in each service description to extract its semantic description.
For instance, supposing that a BP activity label is ”Submission of an invoice”, when applying
the harmonization algorithm, we get the Action ”submit” and the Business Object ”invoice”.
Concerning the SaaS service, we store each solution labels into a dedicated database. For
instance, the SAP concur1 is a SaaS service proposing a set of travel management solutions,
such as expense products, travel products, invoice products, etc. Furthermore, each service
is composed of a set of sub-services. Considering the same SAP concur1 service, it proposes
a set of sub-services such as: ”Broad inventory for global travel booking solutions”, ”Reports
guide informed business decisions”, ”Travel planning that keeps every employee compliant”,
”E-receipts speed up expense claim creation and submission”, etc. These sub services’ labels
are harmonized so as to be syntactically similar to the activities’ ones. To define the semantic
matching activity, three diferent processes are used:
(i) To extend the scope of the BP activities’ label, we used the lexical and the semantic dataset
Wordnet dictionary [
where “nb” represents the number of terms according to the SAAS functionalities ‘i’. The SAAS functionalities having the highest pertinence represents the corresponding SAAS to the BP activities.
The cloud services composition activity is very important in our research context. The input of this activity is the BPMN model subject of the BP to be outsourced to the SaaS services as well as its data model. Its aim is to allocate SaaS services to each SESE fragment in the way that the allocation ensures: • A greater matching between the BP required functionalities and those of the selected
SaaS; • Avoiding outsourcing BP fragments, requiring a considerable amount of external data access, to a set of SaaS services that do not handle data access techniques; • Outsourcing to SaaS services holding role based access control techniques once the outsourced BP activities require to be handled by specific role(s).
From the computational point of view, selecting the most suitable SaaS service for each business process SESE fragment, is a large scale and complex optimization problem that can be resolved by applying methods for combinatorial optimization. To select the optimal set of SaaS services from a variety of SaaS ofers, our proposed resolution approach adopt a heuristic technique namely the genetic algorithm. .
Genetic algorithms (GA) [
A chromosome in the SaaS allocation/composition problem, represents a SaaS ofer that would potentially support the execution of an SESE fragment. Hence, two basic compartments constitute the chromosome. The first one 1 corresponds to the fragment and the second compartment 2 is related to the SaaS ofer supporting the execution of the identified in 1. Is is straightforward mentioning, that the application of the GA is realized multiple times to identify several SaaS that potentially would support the execution of the BP fragments. Instead of representing a chromosome as a set of SESE fragment and the related SaaS ofering, we believe that using the proposed representation is more eficient as it simplifies the fitness function detailed in what follows.
Equation 2 formalizes our objective function for the genetic algorithm. The value given by this function is called the fitness value, and it is used to evaluate the quality of the obtained solution. (4) (5) (6) (7) () =
1 × [1 ×
1 + 2 × [ ×
∑︁ ∑︁ ((, )) =1 =1
∑︁ ∑︁ ((, ))]] =1 =1
1 + 3 × [ + 4 × × 1
] 1 + Where ∑︀4=1 =1 is the weights for each equation in the fitness function.
Line 1 of the fitness function focuses on functional similarities between activities and SaaS, while line 2 compares the data similarity between each SESE and the SaaS managed data. Within an individual, the SaaS may be composite, composed of other services (SAP concur is a composite SaaS), or atomic ( for instance, the sub-product "Travel planning that keeps every employee compliant"). In the case of a composite SaaS, its overall sub-products are evaluated in a way that allows to identify for instance, that 3 (see Fig. X) should be supported by the SaaS "SAP concur".
The line 3 of Equation 2 aims at maximizing the number of role based access control techniques
of the adopted SaaS provider proportionally to the number of roles within the SESE fragment.
Hence, the fitness function evaluates the number of SaaS access control tools provided
by the adopted SaaS provider. is the total number of the most popular and prominent
SaaS access control tools [
1 Finally, refers to the total number of roles in the outsourced SESE fragment. In other words, if the number of roles is important but the adopted SaaS providers do not provide suficient tahcacetsisf controltoo=ls, thanthethenco×mputin1gva1lueisisloowm. iIttteisdsatsratihgehtSfaoarSwparrdovtoidmereonfetirosn, the required access control tools so that the number of roles within the SESE is not important.
The equation in line 5 evaluates the number .
The crossover and the mutation are two genetic operators representing fundamental elements in the GA. For the crossover operator, we use the single point one respecting a predefined probability .
A chromosome is also subject to a mutation operation according to a mutation probability . It is applied to change whether a SESE fragment, or a SaaS service.
The BP owner lacks in general the suficient knowledge and expertise related to cloud and more specifically to SaaS model. As the business process execution is supposed to be entirely supported by a composition of SaaS services, transforming the outsourced business process model along with its activities to the target SaaS composition is of paramount importance for BP owner.
Hence, we propose (i) an extension of the BP meta model in order to be able to illustrate required elements for the SaaS composition namely, the SESE fragment and its relation with other BP elements and the explicit details related to the BP input and output data (ii) a new SaaS meta model allowing to make in evidence how each BP element is handled in SaaS. (iii) We propose finally a model to model transformation from business process model to SaaS model. This transformation is basd on a set of transformation rules we propose but we omit in this section as it is not the focus of the paper.
Even though the business process outsourcing to the cloud is not a recent research area, several
works and propositions are elaborated in order to propose a standard process to safely and
efectively outsource a business process to the cloud [
Recently, SaaS model become among the wise choice of enterprises as a mature alternative
of their in-premise applications or to support the BP execution. [
The objective of our experimental evaluation of the performance of the SaaS Allocator is to evaluate the ability of the GA to identify optimal solution , i. e. the best SESE to outsource and the appropriate SaaS to support its execution
We implemented the genetic algorithm using the Java language. In our experimented implementation of the genetic algorithm, we have fixed its characteristics as depicted in table 1.
To evaluate the performance of the GA, we use a real case related to a BP of travel agency.
The BP is constituted of 7 SESE fragments. It is straightforward to mention that the BP is simplified so the number of SESE fragments is reduced for simplifications sake. The main purpose of the BP is to manage the customer quote through the calculation of the flight, hotel, and car rental quotes.
Concerning SaaS services, we evaluate the overall collected services through the internet whose number surpasses 8000; constituted of services and their sub-products. The SaaS labels refactoring is applied only for the services selected by the SaaS allocator module.
When applying the GA, the best solution identified corresponds to outsourcing the SESE fragment F2 to SAP Concur. This SaaS service ensures managing travels through multiple set of services namely, the expense management, the travel solutions and the invoice automation. The travel solutions encompasses services covering, car rental through the service "select car rental company", Hotel Reservation Services (HRS), travel control and report through the sub-product "TripLink".
In the following, we will present firstly, an example of the calculation related to the semantic similarities between the SESE fragment F2 activities/ data to be outsourced and the adopted SaaS/ data. The required input data for the F2 fragment are: the customer data constituted of his personal information as well as his travel data (car rental, flight and hotel). The output data corresponds to the customer travel quote which is used by the fragments F6 and F7.
Sim_Act(Obtain_carrental,select_car_rental)= 6− 2 2 =0.5 Sim_DataIn(car_rental_data,car_rental_Company)= 6− 2 2 =0.5
For both similarities (activity and data labels), the calculated value is 0.5. The value 2 corresponds to the number of similar terms and 6 is the total number of terms in both labels. Indeed, for the similarity between the activity and the SaaS service label, car_rental is used in both labels. However, as "obtain" and "select" are not synonyms, their similarity value is equal to 0.
The performance evaluation is based on the comparison between the GA results with the Round Robin (RR) and the Greedy ones. For the Round Robin algorithm, the most appropriate solution to adopt is outsourcing the fragment F6 to the hunter1 SaaS. The best solution related to the Greedy algorithm is to outsource F7 to the Concur service. Table 2 illustrates the best solution identified (based on the fitness function) of each algorithm.
According to the provided results of the diferent algorithm, the GA provides better solution in terms of number of roles, number of exchanged data and the precision. The precision computes the similarity of each SESE activity with the provided services of each selected SaaS.
Although the good performance of the SaaS allocator module and its ability to match activities required functionalities with the SaaS ones, we observe some limitations related to the semantic matching module. Indeed, some SaaS providers use specific and not expressive terms to label their services which harden the task of matching the activities to the appropriates services.
This paper proposes a comprehensive framework for business process based SaaS composition. The framework is constituted of four basic modules namely, the BP decomposition, the semantic matching, the SaaS allocator and the SaaS modules. These modules allow to fill the gap between the BP and the SaaS services available on the Internet. Finding out the suitable BP fragment to outsource along with the SaaS to adopt is not a wall-worn task regarding the complexity of the BP structure (i. e. number of activities, number of exchanged data, and number of roles). Hence, we propose, in the SaaS allocator module a genetic algorithm allowing to find good solutions in diferent cases.
Actually, we are working on proposing, in additional to the objective selection of the SaaS to adopt, a subjective evaluation of these cloud services, based on users feedback from social media. Combining objective and subjective evaluation gives a more pertinent solutions.