In requirements engineering for service-oriented systems, stakeholder objectives are described by goal models from which business process models are derived that de ne the required service compositions. A goal-based requirements speci cation should ensure completeness of the goals that need to be achieved as well as the consideration of temporal and logical dependencies between goals. Recent approaches lack the ability to elicitate and specify the stakeholders' knowledge about dependencies between goals that need to be considered in the derivation of business processes. We present Kaos4SOA, an extended goal notation for the explicit speci cation of goal dependencies and brie y describe how existing approaches can be extended to validate the consistency of business processes against these dependencies. The application of our approach is supported by the implemented Kaos4SOA modeling workbench.
Following the paradigm of service-oriented architectures [
The elicitation and analysis of early requirements using goal models has been
addressed comprehensively by existing work [
Goal Model G1 ...
...
OR Goal B
Goal C
Goal D Operation A
Operation B
Operation C
Process Model P1 Process Model P2 Process Model P3
Op A Op B Op C
Op B Op C
Op A Op A Op B Op C
To address this issue, we propose to elicitate the stakeholders' knowledge about logical and temporal dependencies by specifying them explicitly in KAOS goal models. These dependencies are de ned between goals on di erent levels of abstraction and constrain how the operations shall be composed.
The contribution of this paper consists of two parts. We introduce Kaos4SOA,
an extended KAOS goal notation that supports the speci cation of temporal
and logical dependencies between goals [
The remainder of the paper is structured as follows. In the next section the foundations of KAOS goal models are introduced. Section 3 presents the Kaos4SOA approach and discusses the extension of existing approaches to generate veri able business process constraints from Kaos4SOA models. In Section 4, the implemented Kaos4SOA modeling workbench is described. Section 5 discusses related work and nally Section 6 concludes the paper and gives an overview about future work. 2
In this section, we introduce the KAOS modeling approach for specifying goal
models [
An example for a KAOS goal model is illustrated in Figure 2. In order to achieve the main goal Customer order processed, four subgoals need to be ful lled as speci ed by the AND-decomposition. The subgoals Customer data recorded and Payment received are further decomposed into subsubgoals that provide alternatives to achieve the subgoals (OR-decompositions).
The decomposition of Customer data recorded is a special type of decomposition. The circle connecting the goals Account login data entered and Customer pro le loaded indicates an AND-decomposition which itself is part of an ORdecomposition. This means that the goal Customer data recorded is achieved by New Customer account created or by Account login data entered AND Customer pro le loaded.
Using operationalization [
In addition to the AND-/OR-re nement relationship between goals, KAOS
also provides the ability to express con icts between goals. For this purpose, the
Con ictsWith attribute [
In Section 1 we motivated the need for the elicitation and speci cation of goal dependencies to enable their consideration in the derivation of business processes. We propose Kaos4SOA, an extension of KAOS to express these dependencies in a goal model. In addition, we brie y describe the requirements for the automatic generation of veri able constraints. 3.1
Goal Dependency Modeling
As described, KAOS goal models provide the ability to express relationships
between goals (e.g. AND/OR-relation) and pre- and postconditions for atomic
operations. We have analyzed existing research presented in [
First, it is desirable to consider the order in which goals need to be achieved. We argue that even on the abstract level of goals, a requirements engineer is already able to decide about temporal dependencies between goals. In addition, we aim for a more precise de nition of the logical OR-re nement. In the existing KAOS notation the OR-statement is interpreted as an inclusive-OR, i.e. OR( G1 , G2 ) ! G1 _ G2 _ ( G1 ^ G2 ). Currently, it is not possible to de ne conditions to specify when G1 or G2 need to be achieved. For this purpose we introduce annotation capabilities for branches to de ne conditional branches. The proposed extensions of KAOS to address the required dependencies are described in the following.
Order Constraints KAOS goal models do not provide the ability to express dependencies among goals regarding the order in which the goals need to be achieved. For this purpose, we extend the annotation capabilities of KAOS with the additional Order element. Two di erent types of orders are expressible by the proposed extension.
Loose Order describes that a goal G2 needs to be achieved some time later than goal G1. It does not matter, if other goals are achieved meanwhile. For expressing the loose order dependency the attributes Order.Predecessor and Order.Successor are introduced. Figure 3 (a) shows a loose order example: Before the ordered goods are shipped, the goods' availability has to be checked.
Strict Order expresses that the achievement of goal G1 is directly followed by the achievement of goal G2 without other goals achieved in-between. This kind of sequential execution is described by the annotations Order.StrictPredecessor and Order.StrictSuccessor. As an example the strict order The credit worthiness needs to be checked directly before the credit card is debited is shown in Figure 3 (b).
(a) Loose Order Constraint
Order processed AND (b) Strict Order Constraint
Receive credit card payment
AND Availability checked In some cases not only the achievement of single goals is important to be constrained by a condition, instead the achievement of a whole set of goals depends on a condition. Often this can be seen for alternatives, which are all suitable to achieve a desired result, but not each goal sequence is appropriate in each case. An example for conditional branches is depicted in Figure 4. The goal Customer data recorded is decomposed by an OR-re nement. Which alternative subgoal shall be chosen under which condition, can be di erentiated by the new attribute condition which is de ned for both links. In this example, the conditional branch is used to de ne the following policies.
If the customer is ordering for the rst time (is a new customer), he needs to create a new account for recording the customer data.
If the customer already has an account, he needs to login AND the customer pro le needs to be loaded to record the customer data.
Towards the Goal-driven Generation of Business Process Constraints To preserve the consistency between goal models and business process models it is required to evaluate that the derived business processes do not violate the constraints indicated by the temporal and logical dependencies in the Kaos4SOA model.
Kaos4SOA facilitates additional goal annotations expressing the
dependencies, but does not provide formal constraints. To evaluate the consistency of a
business process regarding these dependencies, formalized constraints need to be
generated from the Kaos4SOA goal model. Since the dependencies are de ned
among goals on di erent levels of abstraction they need to re ned iteratively
to the operational level. From the re ned dependencies, business process
quality constraints are generated that express the de ned dependencies by concrete,
veri able constraints. [
The applicability of our approach depends on the ability to create and edit complex goal models and the dependencies among the goals in an e cient way. To provide a tool support for our modeling approach we implemented the Kaos4SOA Workbench. It provides modeling capabilities for standard KAOS modeling elements as well as for the extended Kaos4SOA goal dependencies.
The Kaos4SOA Workbench is realized as a plugin for the Eclipse development environment 1. A metamodel for the extended KAOS notation is de ned by EMF Ecore models using the Eclipse modeling framework (EMF). The concrete syntax for the extended goal models is modeled by the Graphical Modeling Framework (GMF). Based on the de ned GMF models a graphical editor is generated, that enables the creation and editing of Kaos4SOA models. A screenshot of the modeling UI is depicted in Figure 5. 5
The i* framework [
Tropos [
In summary, existing goal oriented approaches provide concepts for expressing and modeling dependencies among goals. Nonetheless, they lack the ability to de ne concrete logical dependencies for branches and do not support any kind of temporal dependencies.
In [
This paper presents Kaos4SOA, an extension of the well-known KAOS approach to elicitate the stakeholders' knowledge about dependencies between goals. Based on an analysis of existing research we identi ed required dependencies and the shortcomings of KAOS for expressing these dependencies. We introduce and de ne new notation elements for logical and temporal dependencies. Finally, we present a tool support that supports the creation of extended KAOS models in a graphical modeling workbench.
The future work of our research is twofold. First we aim to extend our approach with the generation of veri able business process quality constraints, that can be used to evaluate the consistency between goals and derived business processes. Our future work also includes the extension of the Kaos4SOA Workbench with a model checker which allows the integrated, automated veri cation of business process models against de ned constraints.