THE CONVERGENCE OF WORKFLOWS, BUSINESS RULES AND
COMPLEX EVENTS
Defining a Reference Architecture and Approaching Realization Challenges
Markus D
¨
ohring, Lars Karg, Eicke Godehardt and Birgit Zimmermann
SAP Research, Bleichstraße 8, Darmstadt, Germany
Keywords:
Workflow management, Business rules, Complex event processing, Business process management.
Abstract:
For years, research has been devoted to the introduction of flexibility to enterprise information systems. There
are corresponding concepts for mainly three established paradigms: workflow management, business rule
management and complex event processing. It has however been indicated that the integration of the three
paradigms with respect to their meta-models and execution principles yields significant potential for more
efficient and flexible enterprise applications and that there is still a lack in conceptual and technical guidance
for their integration. The contribution of this work is a loosely coupled architecture integrating all three
paradigms. This includes a clear definition of its building blocks together with the main realization challenges.
In this context, an approach for assisting modelers in solving the question which paradigm should be used in
which way for expressing a particular business aspect is presented.
1 INTRODUCTION
The proper integration of workflow management
(WfM), business rule management (BRM) and com-
plex event processing (CEP) paradigms is a major
concern for future enterprise applications. Those are
expected to improve the overall organizational flex-
ibility and performance as well as to provide self-
adaptivity capabilities in terms of tailoring them-
selves to the business needs of an enterprise over time
(Gualtieri and Rymer, 2009). One aspect in this re-
gard is the increased semantic expressiveness of the
meta-models e.g., by combining workflows and busi-
ness rules (Zur Muehlen et al., 2008). However, since
each field has evolved separately and usually targets
a definite class of business problems, there is also a
particular overlap in modeling constructs by means of
the three paradigms. This may lead to confusion re-
garding the question what to model where—and how.
Generally, the provision of a properly defined inte-
grated architecture for WfM, BRM and CEP together
with specific modeling guidelines is still a long way
off (Brett and Gualtieri, 2009). The highlighted gaps
are addressed by this work as follows: In Section 2,
related work on the combination of WfM, BRM and
CEP is discussed. Based on the identified lack of
holistic integration concepts, an architecture for the
proper interplay of the three paradigms is presented
and its building blocks are formalized in section 3.
The section also discusses the main architectural re-
alization challenges in terms of meta-model mapping
and extension as well as the provision of model rec-
ommendations. A basic example illustrates ideas for
their solution in section 4. Section 5 concludes this
work.
2 RELATED WORK AND
MOTIVATION
The work of (Knolmayer et al., 2000) describes
the use of reaction rules as an integration layer be-
tween different workflow modeling languages. They
show how typical workflow constructs, for example
XOR branching, can be realized by Event-Condition-
Action (ECA) rules. The approach is refined in
(Van Eijndhoven et al., 2008), where a methodology
for manually identifying variability points in a work-
flow (parts which are likely to change in the future)
and applying a pattern-based transformation to ECA
rules is proposed. The combination of rule and work-
flow meta-models is targeted by (Milanovic and Ga-
sevic, 2009), who also provide a graphical integration
of industry standard modeling languages for the two
paradigms. (Decker et al., 2007) present a graphical
338
Döhring M., Karg L., Godehardt E. and Zimmermann B. (2010).
THE CONVERGENCE OF WORKFLOWS, BUSINESS RULES AND COMPLEX EVENTS - Defining a Reference Architecture and Approaching
Realization Challenges.
In Proceedings of the 12th International Conference on Enterprise Information Systems - Information Systems Analysis and Specification, pages
338-343
DOI: 10.5220/0002972903380343
Copyright
c
SciTePress
notation for modeling complex events in workflows.
The notation consists in a temporal extension for
BPMN
1
, e.g., introducing precedence relationship
connectors for event nodes. (Paschke and Kozlenkov,
2008) implement whole workflows based on rules, es-
pecially leveraging eventing paradigms. Their efforts
are of more technical nature on the implementation
layer and do not target the special requirements of a
business analyst regarding the overall understandabil-
ity and efficiency of business logic. (Bry et al., 2006)
provide a case study realizing a fictive car rental pro-
cess completely with ECA rules, stating that this bears
potential drawbacks for instance regarding the capa-
bilities of monitoring a business process.
In summary, much related work addresses meta-
model integration issues. Yet confusion is caused by
an improper mixing of the paradigms best suited to
express a particular business problem. This multiplies
with an insufficient conceptual separation of model-
ing paradigms and their implementation within an en-
terprise system. All approaches lack a convenient
methodological support for finding the joint balance
for employing elements from WfM, BRM and CEP.
For this reason, (Zur Muehlen et al., 2008) present a
framework comprising decision criteria for choosing
rule or workflow concepts, however neglecting event-
ing aspects. Those criteria for instance include how
frequently the business aspect changes in its defini-
tion. Even though the criteria are principally mea-
surable to a restricted extent, no formal procedure is
provided on how to determine thresholds and to offer
specific model recommendations. Consequently, an
important lack in research is the provision of the con-
ceptual facility for using the paradigms together and
appropriate guidance to deal with semantic overlaps
of the meta-models.
3 JOINT ARCHITECTURE FOR
WFM, BRM AND CEP
Imagine a business analyst who wants to represent a
customer solvency rating, whose outcome depends on
many interdependent criteria, in an IT system. Would
he models those checks as a sequence of workflow
steps? Or declaratively as business rules? Or would
he rely on events like customer has outstanding pay-
ments > 5000$ for more than 2 months? Could he
even use a combination of all of the paradigms? This
section lays the foundation for such a potential com-
bination in terms of architectural building blocks and
1
See http://www.omg.org/cgi-bin/doc?dtc/09-08-14 for
Business Process Modeling Notation V.2.
corresponding realization challenges, followed by a
formalization of the architecture components and in-
tegration points.
3.1 Building Blocks and Integration
To the best of our knowledge, there is no clear and
holistic definition of a loosely coupled architecture
supporting all three paradigms. Loosely coupled
means that it should be possible e.g., to only employ
a workflow management system and step-by-step ex-
tend it with a rule engine and finally a CEP engine if
needed. Consequently the architecture must specify
the paradigm integration points for industry standard
meta-models like BPMN2 and RIF
2
in a minimal-
invasive way, which means there should be preferably
no changes to the original meta-models. For an event
pattern language (EPL), there is no standard available
yet. Therefore the consideration of proprietary lan-
guages like Esper EPL
3
will be necessary.
A reference architecture that suffices these re-
quirements is presented in figure 1. Its formal defi-
nition is provided in section 3.3. The (meta-)model
layers are separated from the implementation layer
within the architecture to emphasize their partial inde-
pendence. This is because in practice, each paradigm
is realized with a favored set of implementation vari-
ants, as for example petrinet-based workflows, RETE
pattern matching for rule conditions or event-driven
backward chaining for CEP as in (Anicic et al., 2009).
If however the power of the distinct meta-models is
correspondingly constrained, all paradigms could be
implemented within one monolithic software compo-
nent. Neglecting implementation aspects for now, we
describe the features and interplay of the paradigms
in figure 1 from right to left.
CEP captures changes in the state of systems
(eventually from the real-world via sensors) as events
and provides a temporal algebra as well as data-
mining and pattern matching algorithms to detect
events of higher semantic meaning (Luckham, 2002).
These abilities in addition to facilities for handling
and filtering masses of incoming events from dif-
ferent sources are the unique features of CEP. For
our reference architecture, all detected events which
are business relevant according to some conditions
are kept within an event cloud. This term is coined
by (Luckham, 2002) and expresses that events in a
business context usually do not stem from one orga-
nized stream, but from several sources causing a lot of
data noise. One important characteristic of the CEP
2
See http://www.w3.org/TR/rif-core
3
See http://esper.codehaus.org
THE CONVERGENCE OF WORKFLOWS, BUSINESS RULES AND COMPLEX EVENTS - Defining a Reference
Architecture and Approaching Realization Challenges
339
Workflow Model (WF)
Workflow Model (WF)
Business Ruleset (R)
Business Ruleset (R)
Event Cloud (C)
Event Cloud (C)
WFM
paradigm
Model
Implementation
layer
BRM
CEP
e.g. event-driven
backward chaining
e.g. RETEe.g. Token Passing
Business Logic (T)
Conditions (P)
+ Event Algebra
Business Logic (T,D)
Conditions (P)
Workflow Language
(e.g. BPMN2)
Workflow Language
(e.g. BPMN2)
Rule Definitions
(e.g. RIF)
Rule Definitions
(e.g. RIF)
Event Patterns
(e.g. Esper EPL)
Event Patterns
(e.g. Esper EPL)
Model Integration (Transformation or Weaving)
Meta-
Model
Model Guidelines and Recommendations
invoke (e, g)
Partly Shared Task Repository (T)
Partly Shared Data Model (S)
Sensor
Events
raise (e)
triggger / raise (e)
trigger (e)
Figure 1: WfM-BRM-CEP integration architecture.
paradigm is, that it is not responsible for conduct-
ing business logic. It is rather a mere event cruncher.
Events from the cloud can be used as signals to trig-
ger business rules or for catching event nodes within
workflows.
BRM aims at enabling the business domain expert
to express statements that define or constrain aspects
of the business (Hay and Healy, 2000). Being spec-
ified in a modular and declarative way as relatively
simple and small ECA rules, the realization of com-
plex decision and action logic can be achieved by the
collectivity of a ruleset. In summary, the strength of
business rules is their easy comprehensibility and di-
rect separate changeability. Within our reference ar-
chitecture, business rules have at most one event at-
tached as a direct trigger. If constraints on multiple
events or temporal constraints need to be specified,
this should be achieved in the CEP environment yield-
ing a new single complex event. This has the advan-
tage of disburdening the business analyst (i.e. the rule
expert in this case) as well as the BRM engine from
having to deal with complex event algebra. The ac-
tion part of an ECA rule may contain manipulations
to the local data context or task executions, which in
turn can lead to the detection of new complex events.
Business rules in our architecture are of twofold rele-
vance: They can be explicitly invoked by a WfMS as
a decision service or they can be triggered by events
from the event cloud providing reacting resp. excep-
tion handling mechanisms.
WfM aims at the IT-based specification and coor-
dination of tasks as units of work within an enterprise
and their allowed execution order to jointly realize a
business goal (Leymann and Roller, 1999). In prac-
tice, a workflow model is implemented by IT experts
in cooperation with domain experts and remains rel-
atively stable over a longer period of time. However,
in some niche areas, also adaptive WfMS as presented
by (Dadam and Reichert, 2009) have evolved. Such
WfMS allow for more flexibility in terms of control
and data flow changes at design- and runtime. The
strength of workflows lies in the overall comprehen-
sibility and traceability of business logic. A workflow
as part of our architecture may receive and commu-
nicate events with the event cloud. To be able to in-
tegrate business rule conditions with workflow task
execution, we assume that at least the finishing of
a task instance constitutes a business relevant event
comprised by the event cloud. As such, it can be
reused amongst others for triggering business rules.
The branching behavior of gateways can be realized
by invoking business rules.
For an efficient integration of the three paradigms,
there must be a partly shared data model for all
paradigms, e.g., a customer ID must be uniquely de-
fined. Furthermore, there must be a shared task repos-
itory for use in rules and workflows, i.e., a predefined
task can be started within a WfMS as well as the ac-
tion part of a rule. The white arrows in figure 1 sym-
bolize the architectural integration points. The two
horizontal continuous arrows concern the main scope
of this work. The one on the meta-model layer rep-
resents the ability to model distributed business logic
combining all three paradigms (Section 3.2.1). The
one on the model layer relates to the provision of de-
cision and optimization support (Section 3.2.2).
3.2 Realization Challenges
3.2.1 Challenge #1 - Integration of the
meta-Models
Each distinct paradigm, WfM, BRM and CEP has
evolved from specific business domains and exist-
ing meta-models have been designed accordingly. It
is desirable to provide a loose coupling and there-
fore to allow for a stepwise extension of the sys-
tem landscape. Consequently the meta-model integra-
tion should be achieved preferably leaving their origi-
nal definition (almost) intact, using either weaving or
mapping concepts:
Meta-model Weaving. One option to realize the in-
tegration of meta-models without changing or reduc-
ing their original semantics is to define additional
bridge meta-elements. The resulting overall meta-
model is a superset of the original distinct meta-
models regarding its semantic expressiveness. For in-
stance, a new integrated meta-element could be de-
fined using multiple inheritance from different meta-
model packages and subsequently be enriched with
additional properties. An example would be a rule-
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
340
gateway with joint semantics from BPMN and a rule
language.
Meta-model Mapping. Assuming that not all parts
of the meta-models can be reasonably combined by
model extension, it is required to be able to map par-
ticular constructs from one paradigm to those of an-
other paradigm. For instance, if a company has em-
ployed only a rule-based information system in the
past, it should be possible to smoothly migrate some
of the rules that in fact realize process logic to an
explicit workflow representation. The resulting over-
all meta-model is equal to the original distinct meta-
models regarding its semantic expressiveness. An ex-
ample would be the expression of workflow-patterns
via ECA rules.
3.2.2 Challenge #2 - Modeling Guidance
As model mappings will presumably play a chief part
within our meta-model integration efforts, we face the
problem of a semantic redundancy of some parts of
the overall meta-model. That means, when employ-
ing more than one paradigm at once within an enter-
prise information system, it might not be a-priori clear
which construct from what paradigm should be used
to model a particular business problem. Therefore,
we propose to continue and more substantiate work
like (Zur Muehlen et al., 2008) for providing specific
modeling guidelines in terms of grounding them on
solid comprehensible metrics. These can be subdi-
vided into design-time metrics relying on static model
aspects and log-based execution analysis metrics.
3.3 Formal Definitions
We formalize the architecture components mainly for
a definition of the paradigm intersections, e.g., where
modeling elements can be commonly reused. The
presented notation is also employed in section 4 for
sketching the solution approach.
Definition 1 (Event Cloud). Let C be an event cloud
consisting of events e. An event e can be represented
as a tuple e = (s, r,
K
,P), where:
s is a stamp relating to a fixed point in time when
the event occurred or has been detected
r represents a boolean variable whether this event
is business relevant in terms of affecting the busi-
ness logic or not
K
as a potentially empty causality vector contains
all events that caused the detection of e
P M, M := A × V is a set which contains
attribute-value mappings at time s, which are ad-
ditional prerequisites constraining the business
data context the event may occur in.
For the definition of a workflow model we adopt for-
malisms from (Yongchareon et al., 2010), especially
considering the BPMN2 intermediate event construct.
Definition 2 (Workflow Model). A workflow model
is a set of connected tasks, events and gateways in
terms of an extended directed graph. It can be repre-
sented as a tuple W F =
O,T,E,G,F, T
E
,X
, where
:
O represents a set of workflow nodes, where:
T O is a set of tasks. For each t T,e
E
|
r
e
= true,
~
K =
/
0 as an event indicating the
proper finishing of a task instance.
E O C as a set of events E T =
/
0 with a
function type
event
: E
{
start, intermediate
throw
,intermediate
catch
,end
}
G O is a set of gateways G E =
/
0 G T =
/
0
with a function type
gateway
: E
{
or, xor, and
}
.
F O ×O represents the directed control flow re-
lations, implying f (o
i
,o
j
) 6= f (o
j
,o
i
), i 6= j.
T
E
E ×T is a set of attachments of intermediate
events on tasks.
X M, M := A ×V is the valid data context in
terms of workflow variables and possible values.
Definition 3 (Business Rule). A business rule r from
a ruleset R in the ECA format is defined as a triple
{
e E e =
/
0,P
0
M,
~
A := a
i
S
T
i
O
i
O
T
i
D
,
where
~
A can be interpreted as an ordered list of ei-
ther task executions or data context change opera-
tions from the set D as a function d D : m
1
M
m
2
M, m
1
6= m
2
Implications for Architecture Bindings.
g
|
[
|
( f F
|
f
1
= g )
|
> 1 type
gateway
(g)
6= and], ρ R specifying the split behavior.
The shared data model implies that S M, M :=
A ×V commonly used in C, R, O.
The shared task model implies:
[
T
i
O
i
O
T
i
[
T
j
R
j
R
T
j
6=
/
0
4 APPROACHING THE
CHALLENGES: EXAMPLE
In this section, we concretize some suggestions for
approaching the architecture realization challenges.
For this purposes, we build upon our aforementioned
THE CONVERGENCE OF WORKFLOWS, BUSINESS RULES AND COMPLEX EVENTS - Defining a Reference
Architecture and Approaching Realization Challenges
341
Record Order
Check
Technical
Feasibility
Check
Economic
Feasibility
Check Credit
Rating
Check
Product
Availability
Determine
Alternative
Offer
Send Offer
Reject Order
Accept Order
Close Case
CFC=15; CFC
rel
=1,875
available
unavailable
bad
good
infeasible
CFC=0; CFC
rel
=0
CFC=13; CFC
rel
=2,16
infeasible
feasible
feasible
profit margin < 5%
Figure 2: Workflow with decomposition and metrics.
Record Order
Determine
Alternative
Offer
Send Offer
Close Case
Decide on Request
(a) Abstracted workflow.
1. ON
technical_check_finished
credit_check_finished
availability_check_finished
economic_check_finished
RAISE
checks_finished
2. ON
credit_check_finished
IF
creditCheck.result=failed
rejectTask.started=false
START
reject
(b) Workflow parts mapped to
Events&Rules.
Figure 3: Combination of workflow and rules.
example of a business analyst who wants to model
complex checks for a customer and present an order
processing workflow extended from (Seel et al., 2006)
in figure 2. After a customer order has been received,
the default continuation is to execute four checks, de-
pending on which the order is rejected or accepted. If
the profit margin of the ordered product is not high
enough, an alternative offer is sent to the customer
without accepting or rejecting the actual order. The
decision logic encoded in the workflow diagram is
quite complex and hard to capture. It therefore makes
sense to outsource a part of the flow decision logic to
a rule base.
To address challenge #2, design-time model com-
plexity metrics may contribute to the identification
of model optimization potentials. For our example,
we are interested in complex parts of the workflow
which are hard to capture via flow diagrams. There-
fore the workflow is first decomposed by applying
mechanisms as presented in (Polyvyanyy et al.,
2009). The procedure identifies workflow segments
as φ = (T
0
T, G
0
G,F
0
F). Characteristics like
intermediate events as permitted by definition 2 are
not contained within the example, but are however
considered in workflow segmentation work by
(Yongchareon et al., 2010). The detected segments
are marked with a dashed line in figure 2. To measure
the complexity of a workflow segment, (Cardoso,
2006) provides an intuitive and empirically validated
metric: the control-flow-complexity (CFC), which
is based on the number of mental states a modeler
has to consider when modeling the workflow caused
by split gateway semantics. The metric can be set
in relation to the number of task nodes within a
workflow segment to express the “density” of deci-
sion complexity for φ as CFC
rel
= CFC (φ)
T
0
φ
.
The corresponding values for CFC and CFC
rel
are indicated in the upper left corner of the work-
flow segments. Having identified segments in the
workflow most appropriate for being outsourced as
rules, they can be transformed into the rule format
provided by definition 3 (relating to challenge #1)
by applying pattern-based transformations like
{
f
1
(t
1
,g
1
), f
2
(g
1
,g
2
), f
3
(g
2
,t
2
)
}
,type
gateway
(g
1
)
= XOR, type
gateway
(g
2
) = AND r =
t
1
f inish, P
f
2
, start (t
2
)
. As such, rules as
presented in extracts in figure 3(b) can be con-
structed. As shown in figure 3(a), the transformed
segment in the workflow is replaced by a placeholder
task, with the intermediate event checks f inished
attached for continuing the control flow as soon as a
decision can be drawn from the checks.
As an intermediate conclusion, the benefits of the
“outsourced” rules as provided by our approach are
manifold. There is a potentially better understand-
ability of single aspects of the decision logic (not
necessarily the whole logic!). One can see for ex-
ample immediately that a failed credit check directly
leads to an order rejection. Furthermore, process
extensions and exception handling mechanisms can
be added more conveniently. A complex event caused
by the customer overdrawing his account for more
than 2 months could raise a reject finished event for
all order processes of the customer and cause the
untimely closing of the instances.
5 CONCLUSIONS
This paper presented an integrated reference archi-
tecture for the combined modeling of business logic
with workflows, business rules and complex events.
The provision of an integration layer for the different
meta-models as well as the supply of modeling guide-
lines were identified as the two main realization chal-
lenges for the architecture. To provide a starting point
on how the challenges can be tackled, a procedure
to identify complex decision logic within a workflow
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
342
and outsource it to rules was presented. The proce-
dure is based on design-time complexity metrics for
workflows and pattern-based transformations. As fu-
ture work, existing standards for the three paradigms
like RIF and BPMN2 are to be examined to deter-
mine their meta-model compatibility. Furthermore, it
is required to conduct a study on which granularity
for each paradigm is appreciated by business analysts
and how this can be related to metrics.
ACKNOWLEDGEMENTS
Thanks to Anis Charfi and Todor Stoitsev for helpful
comments. The work presented in this paper was de-
veloped in the context of the project Allianz Digitaler
Warenfluss (ADiWa) that is funded by the German
Federal Ministry of Education and Research. Support
code: 01IA08006.
REFERENCES
Anicic, D., Fodor, P., Stuhmer, R., and Stojanovic, N.
(2009). Event-Driven Approach for Logic-Based
Complex Event Processing. In CSE 2009, Vancouver,
Canada. IEEE.
Brett, C. and Gualtieri, M. (2009). Must You Choose Be-
tween Business Rules And Complex Event Processing
Platforms? Forrester.
Bry, F., Eckert, M., Patranjan, P., and Romanenko, I. (2006).
Realizing business processes with ECA rules. In PP-
SWR 2006, volume 4187, pages 48–62, Budva, Mon-
tenegro. Springer.
Cardoso, J. (2006). Process control-flow complexity metric:
An empirical validation. In SCC ’06, pages 167–173,
Chicago, USA. IEEE.
Dadam, P. and Reichert, M. (2009). The ADEPT Project.
Computer Science - R&D, 23(2):81–97.
Decker, G., Grosskopf, A., and Barros, A. (2007). A graph-
ical notation for modeling complex events in business
processes. EDOC ’07, page 2736.
Gualtieri, M. and Rymer, J. R. (2009). The Forrester
Wave: Complex Event Processing (CEP) Platforms,
Q3 2009. Forrester.
Hay, D. and Healy, K. (2000). Defining business rules -
what are they really. Business Rules Group.
Knolmayer, G., Endl, R., and Pfahrer, M. (2000). Modeling
Processes and Workflows by Business Rules. In BPM,
pages 16–29. Springer.
Leymann, F. and Roller, D. (1999). Production Workflow -
Concepts and Techniques. Prentice Hall.
Luckham, D. (2002). The Power of Events. Addison-Wesley
Professional.
Milanovic, M. and Gasevic, D. (2009). Towards a Lan-
guage for Rule-Enhanced Business Process Modeling.
In EDOC ’09.
Paschke, A. and Kozlenkov, A. (2008). A Rule-based Mid-
dleware for Business Process Execution. In Multikon-
ferenz Wirtschaftsinformatik.
Polyvyanyy, A., Smirnov, S., and Weske, M. (2009). The
triconnected abstraction of process models. In BPM
’09, pages 229–244, Ulm, Germany. Springer.
Seel, C., Simoin, B., and Werth, D. (2006). Business
Rule-enabled Process Modelling. In e-Challenges,
Barcelona, Spain.
Van Eijndhoven, T., Iacob, M.-E., and Ponisio, M. L.
(2008). Achieving Business Process Flexibility with
Business Rules. In EDOC ’08, pages 95–104, Mu-
nich, Germany. IEEE.
Yongchareon, S., Liu, C., Zhao, X., and Kowalkiewicz, M.
(2010). BPMN Process Views Construction. In DAS-
FAA ’10, Tsukuba, Jaban.
Zur Muehlen, M., Indulska, M., and Kittel, K. (2008). To-
wards Integrated Modeling of Business Processes and
Business Rules. In 19th Australasian Conference on
Information Systems.
THE CONVERGENCE OF WORKFLOWS, BUSINESS RULES AND COMPLEX EVENTS - Defining a Reference
Architecture and Approaching Realization Challenges
343