USING DISTRIBUTED CSPs TO MODEL BUSINESS PROCESSES

AGREEMENT IN SOFTWARE MULTIPROCESS

Luisa Parody, Mar´ıa Teresa G´omez-L´opez, Rafael M. Gasca and Diana Borrego

Department of Computer Languages and Systems, University of Seville, Seville, Spain

Keywords:

Business process management, Distributed constraint satisfaction problem, Web services.

Abstract:

A business process consists of a set of activities which are performed in a coordination way to obtain an

objective. Sometimes the deﬁnition of this objective using only a classic business processes management is

not possible. When the choreography of the processes cannot be deﬁned with a combination of tasks using

sequences, conditions, ’xor’, ’or’ and ’split’ control ﬂow patterns, another representation and solution are

necessary to be used. This problem makes difﬁcult the decision making in software management projects.

In this paper a way to describe a process agreement is described where the execution and the number of

tasks execution order of the Web Services cannot be deﬁned. As a case study, the resource distribution in

a multiproject development environment is used. In this case, the processes have to achieve an agreement

in function of the business rules that relate the processes. In order to achieve this objective, the Distributed

Constraint Satisfaction Problems are used to model and solve this type of problems.

1 INTRODUCTION

Business Process Management (BPM) has received

considerable attention since it permits to combine di-

fferent tasks to obtain a common objective. A busi-

ness process consists of a set of activities that are per-

formed in coordination in an organizational and tech-

nical environment (Weske, 2007). BPM includes con-

cepts, methods, and techniques to support the design,

administration, conﬁguration, enactment, and analy-

sis of business processes. The bases of BPM are

the explicit representation of business processes with

their activities and the execution constraints between

them. However, the aspect analyzed in this paper is

when the execution order and the constraint relation

between the processes cannot be described with a tra-

ditional business process model in an statical way.

The execution order of the activities will be deﬁned

in function of the data ﬂow for the different instances.

For example if the outputs of a business process A af-

fect to another process B, and the output of process

B affects to business A and there is no order between

them. In this case, the processes have to achieve an

agreement in function of business rules that relate the

processes.They have to share a resource to obtain the

maximum beneﬁt.

The combination of processes may cause the use

of services with private and public information, in

such a way that the decisions cannot be made in an ex-

ternal or centralized way to the processes. That is, the

information is distributed in different nodes although

all of them have to work together. This type of Busi-

ness Process behavior has some aspects in common

with DisCSPs hence we think that some of the advan-

tages of DisCSPs can be adapted to solve the Business

Process Agreement.

A company specialized in Custom Software De-

velopment Products tends to work in a multiproject

environment. The complexity inherent in a software

development management increases in the presence

of several concurrent projects which are performed

by the same organization that must allocate resources

between them. (Navascu´es Fern´andez-Victorio and

Ramos Rom´an, 2008) presents a basic scheme of

business process to this type of companies. The orga-

nization plans the production of a project by means of

the assignment of resources based on its capacity, the

agreements acquired with clients and the global work-

load. In other words, based on the business agree-

ments for each project. In addition, the company has

an internal control system to manage the execution

and change the assignment to deal with internal or ex-

ternal unexpected events in the projects. It is difﬁcult

to determine when the process will end since all its

features cannot be known ”a priori”.

Moreover, the company has to establish the dead-

434

Parody L., Teresa Gómez-López M., M. Gasca R. and Borrego D..

USING DISTRIBUTED CSPs TO MODEL BUSINESS PROCESSES AGREEMENT IN SOFTWARE MULTIPROCESS.

DOI: 10.5220/0003180904340438

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 434-438

ISBN: 978-989-8425-41-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

lines of the projects with each client since the pla-

nning is very important to the success or failure of the

project. The resources management is crucial since

the concurrent projects compete for the resources

existing within the company. To reach a solution, they

have to achieve an agreement ensuring their interests

and giving in some cases. Each project has its own

contract with client established by the business rules,

where time delay and ﬁnal cost of the project are re-

lated.

The structure of the paper is as follows: Sec-

tion 2 includes some relevant proposals. Section 3

presents some basic concepts related to DisCSP and

its Asynchronous Backtracking Algorithm. Section

4 describes how the algorithms for DisCSPs can be

adapted to our problem. Section 5 shows an example

of combination of Web Services and describes how

our proposal is applied to the example. Section 6

depicts the experimental results. Finally conclusions

and future work are presented.

2 RELATED WORKS

The graphical Standard for modeling business pro-

cesses is Business Process Modeling Notation

(BPMN) proposed by OMG. However, BPMN stan-

dard is not powerful enough since, among other is-

sues, there is an important need of many adaptive pro-

cesses for concurrentactivities to respect coordination

constraints. These require that concurrent activities

coordinate their behaviors in response to exogenous

events. In order to represent the choreography of the

processes in a business process, several graphical lan-

guages have been developed. The most important is

the standard BMPN (Weske, 2007), that can be used

for large number of problems (Wolter and Schaad,

2007).

3 DISTRIBUTED CONSTRAINT

SATISFACTION PROBLEM

In (Yokoo and Hirayama, 2000), Makoto et al. pre-

sented a Distributed Constraint Satisfaction Problem

(DisCSP) as a general formalism for dealing with

problems in multi-agent systems. A DisCSP is a

CSP where the set of variables and constraints of

the problem are distributed between a set of agents.

These agents are in charge to solve their own sub-

problem and must coordinate themselves with the rest

of agents to reach a solution to the global problem

(Abril L´opez et al., 2007).

In order to improve the time to ﬁnd the best so-

lution, the paper of Redouane Ezzahir et al. (Ezza-

hir et al., 2008) proposes algorithms for solving Dis-

tributed Constraint Optimization Problems (DCOP).

3.1 Backtracking Algorithms to Solve

DisCSPs

There are different proposals to model and solve

DisCSP:

Centralized Backtracking selects a leader agent

among all agents who gather all the information about

variables, their domains, and their constraints (Yokoo

and Hirayama, 2000). If the knowledge about the pro-

blem can be gathered into a single agent, this agent

can solve the problem alone by means of the use of

normal centralized constraint satisfaction algorithms.

Asynchronous algorithms are characterized by the

fact that all agents are active in parallel and only coor-

dinate as needed to ensure consistency if their varia-

bles are involved in the constraints.

In Asynchronous Backtracking (ABT), the prio-

rity order of variables/agents is determined, and each

agent communicates its tentative value assignment

to its neighboring agent via ’ok?’ messages. Each

agent maintains the current value assignment of other

agents from its viewpoint. An agent changes its

assignment if its current value assignment is not con-

sistent with the assignment of higher priority agents.

If there exists no value that is consistent with the

higher priority agents, the agent generates a newcons-

traint (called a nogood), and communicates the no-

good to a higher priority agent, thus the higher prio-

rity agent changes its value making backtracking.

In (Yokoo and Hirayama, 2000) the soundness,

completeness and termination of ABT algorithm is

proved.

3.1.1 Heuristics for Variable Ordering

There are different forms to order the variables to es-

tablish the priority: (a) The highest priority is assig-

ned to the variable that participates in more cons-

traints, called variable degree. There are different

possibilities to establish the next highest priority va-

riable: (a.1) The next variable that participates in

more constraints, without considering the agent who

it belongs to. (a.2) The next variable that participates

in more constraints and belongs to the same agent that

the ﬁrst one. (b) The highest priority is assigned to

the variables that belong to the fastest agent. (c) The

priority is assigned according to the problem speciﬁ-

cation.

USING DISTRIBUTED CSPs TO MODEL BUSINESS PROCESSES AGREEMENT IN SOFTWARE MULTIPROCESS

435

4 USING DISCSP TO MODEL

BUSINESS PROCESS

AGREEMENTS

Business Process Execution Language for Web Ser-

vices (WS-BPEL) is emerging as the prominent lan-

guage for Modeling Executable Business Processes.

This is due to WS-BPEL provides an appropriate

set of constructs to design a process in an intuitive

way. One of its features is the ability to synchro-

nize concurrent ﬂows within a process by specifying

control links between activities. There could be ad-

ditional types of constraints between concurrent ac-

tivities. The constraint that involves the variables of

the different agents, implying their equality, requires

actions in response to events in order to be coordi-

nated. Such types of constraints between concurrent

activities are called coordination constraints (Wu and

Doshi, 2008). These constraints must adapt to events

to preserve their optimality. Unfortunately, coordi-

nation constraints are not natively supported by WS-

BPEL, its support for designing ﬂexible processes is

limited.

On the other hand, the aim of these types of pro-

blems is to obtain a common objective, which de-

pends on the business agreements of each agent, the

client and on the contract established through busi-

ness rules.

The activitiesthat need this type of coordinationto

obtain the overall goal have some aspects in common

with the features of DisCSPs discussed in Section

Establish the

order of variables

Initialize

Variables

Update

STAGE

Calculate

OBJECTIVE

Update

SOL

Update

SOL_OP

Check end

iteration

Yes

Backtrack

Yes

Check end

iteration

Yes

Legend:

Parallel - And

Exclusive - Xor

Sequence Flow

Start Event

End Event

Activity

Sequence Flow Looping

Figure 1: Backtrack Business Process.

3. In this paper, our proposal adapts DisCSPs to the

agreement problems between tasks to obtain the ob-

jective. As the algorithm based on the algorithms pre-

sented in Section 3.1, we have designed a new algo-

rithm to make Business Processes. It can be classiﬁed

as a mixture of Centralized and Asynchronous Back-

tracking. It is centralized since the Business Process

has the knowledge about the values of the variables,

and it is asynchronous since all the agents are active

in parallel and communicate their consistencies or in-

consistencies asynchronously.Thekey is that the algo-

rithm invokes to an activity that calculates the objec-

tive. Then, we translate this algorithm into a Business

Process (Figure 1).

The most important activity in the BP (Figure 1) is

Calculate OBJECTIVE activity, where each problem

models its way to obtain the overall goal.

5 MOTIVATING EXAMPLE

As an example to illustrate our proposal, a Web Ser-

vice to perform the assignment of people to tasks pre-

sented in the introduction (Section 1) is used. During

the development of a software product, different tasks

or activities take place, giving rise to the software life-

cycle.

Typically, a lifecycle model addresses the phases

of a software project: requirements, design, imple-

mentation, integration, testing, operations and main-

tenance. In our example, the company has to deal

with the performing of the implementation phase of

some different projects concurrently. The staff is

made up of different number of workers on each di-

fferent professional category, who must be assigned

properly to perform the tasks of these projects. The

proﬁts obtained by the company depends on the delay

of project. The fulﬁllment of the delivery time agreed

with the customer provides the maximum proﬁts for

each project. An optimal staff distribution will do the

sum of proﬁts as maximum as possible. The calcula-

tion of the proﬁts of each project is internally made

within each project separately, so that it is necessary

to get an agreement to achieve the overall goal. For

example, we have three projects. If two workers are

assigned to project 1, it implies that the proﬁt will be

1 million euros and for project 2, the proﬁt will be

0.4 million. But, if these two workers are assigned

to project 2, the proﬁts will be 0.8 and 0.7 million

for project 1 and 2 respectively, so this is the best

assignment to obtain the maximum global proﬁt.

The proﬁts that the company obtains by each

project depend on the business agreement between

each project manager, the project’s client, and by the

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

436

contract established through business rules. The cal-

culation of the proﬁts of each project is performed in a

different Web Service. The combination of those cal-

culations will provide the overall proﬁt to maximize.

5.1 Backtracking Algorithm applied to

the Motivating Example

In the particular case of the Software Management

Project, there are three different projects that work

concurrently and need some kind of staff organiza-

tion. Although these projects belong to the same com-

pany, they do not have to reach the same agreements

with clients or the internal contracts. For security and

privacy reasons, they share only the necessary infor-

mation to achieve the overall goal. Each project will

correspond to an agent. These agents have to agree to

maximize the proﬁt of the company, trying to ﬁnish

their projects on time with the available staff.

In this section, the proposal presented in Section

4 is applied to the problem of Software Management

Projects.

Considering the cases studied in Section 3.1.1 and

the presented problem, all the variables are involved

in the same number of constraints and there are no

signiﬁcant differencesin the speed of response time of

the different agents. In this case, to establish the prio-

rity order it is necessary to take into account the over-

all goal of the problem: to maximize the overallproﬁt.

The variables which belong to the project which gives

the greatest proﬁt will have the highest priority. With

this criterion, the algorithm may ﬁnd a good solution

quickly and make more efﬁcient bounds.

The most important activity in the BP (Figure 1)

was the Calculate OBJECTIVE activity, in the mo-

tivating problem this activity is Calculate PROFIT.

In this activity there are such parallel ﬂows belong-

ing to each agent. Each ﬂow has two activities: (a)

Solve variables of the agents: it uses the constraints

that the agent has associated to assign consistent val-

ues to variables. (b) Establish the proﬁt: it calls the

Web Service which calculates the proﬁt with the val-

ues assigned to variables in the previous activity.

6 EXPERIMENTAL RESULTS

Table 1 shows the staff distribution given the follow-

ing input: 5, 10 and 6 members of staff with cate-

gories 1, 2 and 3 respectively. The proﬁt obtained

with this distribution is 840500 euros.

Table 1: Results.

Variables P1 P2 P3

num. of T1 4 0 1

num. of T2 8 1 1

num. of T3 1 5 -

7 CONCLUSIONS AND FUTURE

WORK

In this work, a Business Process that adapt the Back-

tracking Algorithm for DisCSP to Business Process is

presented. This proposal arises from the need to coor-

dinate different activities that belong to a Business

Process and work concurrently sharing resources.

Thanks to this adaptation, the activities achieve the

coordination to obtain an overall goal, which, in the

example presented in this paper, is the software multi-

project proﬁt.

One of the main problems in the search of solu-

tions is the size of the search space. A bound could

be established to avoid calling agents and calculate

the proﬁt if we know that we are not going to get a

better result. We could obtain the bound from: (i) the

search space that is already done and (ii) the ”post-

morten” memory (Navascu´es Fern´andez-Victorio and

Ramos Rom´an, 2008) of previous projects.

ACKNOWLEDGEMENTS

This work has been partially funded by the Junta

de Andaluc´ıa by means of la Consejer´ıa de Inno-

vaci´on, Ciencia y Empresa (P08-TIC-04095) and by

the Ministry of Science and Technology of Spain

(TIN2009- 13714) and the European Regional Deve-

lopment Fund (ERDF/FEDER).

REFERENCES

Abril L´opez, M., Barber Sanchs, F., and Salido Gregorio,

M. A. (2007). Particionamiento y Resoluci´on Dis-

tribuida Multivariable de Problemas de Satisfacci´on

de Restricciones. PhD thesis, Universidad Politcnica

de Valencia, Valencia.

Ezzahir, R., Bessiere, C., Benelallam, I., Bouyakhf, H., and

Bela¨ıssaoui, M. (2008). Dynamic backtracking for

distributed constraint optimization. In ECAI, pages

901–902.

Navascu´es Fern´andez-Victorio, J. and Ramos Rom´an, I.

(2008). Un modelo para la simulaci´on h´ıbrida de la

producci´on de software a medida en un entorno mul-

tiproyecto. PhD thesis, Universidad de Sevilla.

USING DISTRIBUTED CSPs TO MODEL BUSINESS PROCESSES AGREEMENT IN SOFTWARE MULTIPROCESS

437

Weske, M. (2007). Business Process Management: Con-

cepts, Languages, Architectures. Springer.

Wolter, C. and Schaad, A. (2007). Modeling of task-based

authorization constraints in bpmn. In BPM, pages 64–

79.

Wu, Y. and Doshi, P. (2008). Making bpel ﬂexible - adapt-

ing in the context of coordination constraints using

ws-bpel. pages 423–430.

Yokoo, M. and Hirayama, K. (2000). Algorithms for dis-

tributed constraint satisfaction: A review. In AAAI,

volume 3, pages 198–212.

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

438