Contributed on October 6, 2011

7.4K

2.1K

Frank Leymann @ BPMN 2010

Transcript

BPEL vs. BPMN 2.0:
Should You Care?Should You Care?
(2nd I tl W k h BPMN 2010 P t d G O t b 13 2010)(2nd Intl. Workshop BPMN 2010, Potsdam, Germany; October 13, 2010)
Prof. Dr. Frank Leymann
Institute of Architecture of Application Systems
University of Stuttgart Phone +49‐711‐7816 470
Universitätsstr. 38
70569 Stuttgart
Germany
Fax +49‐711‐7816 472 
e‐mail Leymann@iaas.
uni‐stuttgart.de
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 2
Brief History of Workflow Technology (no scientific rigor intended)
2010
Modeling/Execution Convergence
Composites as a Service (CaaS)
2005
Graphical Modeling
g/ g
2000
WF A i
Web Service Composition
Graphical Modeling
WF in eScience
1995 Production WF
WF Automation
WF St d di ti B i
WF use in EAI
1990 BPR
WF‐based Apps
WF Standardization Begins
D R i /C P i
People Facing
1985
1990
Transactional WF
Mega Programming
© Frank Leymann 3
Document Routing/Case Processing1985
Office Automation, Forms Processing,…
g g g
Workflow Language Genealogy
BPELJ ?…?BPEL4People BPEL-SPE BPMN 2.0
(OMG)
BPEL
WS-CDL
(W3C)
BPMN 1.x
(OMG)
Petri-Nets
BPEL
(IBM, Microsoft, BEA
OASIS)
GSFL
GPEL
BPEL4Chor
WSFLXLANG
(IBM)(Microsoft)
YAWLGSFL …
(Research Groups)
FDML W/XPDL BPML WSCI BPSS
(IBM) (WfMC) (bpmi org) (SUN, SAP, (OASIS
Pi-Calculus++
(Microsoft) (IBM) (WfMC) (bpmi.org) (SUN, SAP,
Oracle)
(OASIS
ebXML)
(Microsoft)
-Calculus
© Frank Leymann 4
FDL
(IBM)PM-Graphs
Process Modeling is (has been?) Layered
Conceptual Process Model
BPMN
TransformProcess Model
Logical Process Model
BPEL
Deploy
Process Engine
© Frank Leymann 5
Process Modeling is (has been?) Layered
Conceptual Process Model
BPMN
Transform
Logical Process Model
BPEL
Deploy
Process Engine
© Frank Leymann 6
BPM Lifecycle: Where BPMN Fits
Modeling
KPIs
g
IT RefinementAnalysis
DeploymentMonitoring
New in 2.0!
(but this is optional)
© Frank Leymann 7
ExecutionExecution
BPMN 2.0 vs. BPMN 1.1
BPMN 2.0 is based on BPMN 1.1 but additionally 
contains:contains: 
An XML‐based interchange format for BPMN models
Clean operational semantics and metamodelClean operational semantics and metamodel 
(connecting BPMN to OMGs MDA efforts)
The metamodel is one of the most important changes, hence theThe metamodel is one of the most important changes, hence the 
new name (see below)
The graphical notation is mostly unchanged
some new event types, choreography, collaboration,…
New BPMN “Level”: executable BPMNNew BPMN  Level : executable BPMN
BPMN in 1.1: “Business Process Modeling Notation”
© Frank Leymann 8
BPMN in 2.0: “Business Process Model and Notation”
Overview On All Graphical Artifacts
© Frank Leymann 9
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 10
The Key Aspect of BPEL
BPEL exists and is supported by most middleware vendors!
Combination of two modeling paradigms
Avoided split of WF product market/industriesAvoided split of WF product market/industries
The runtime standard
Well defined syntax & operational semanticsWell defined syntax & operational semantics
(Certain degree of) portability across vendors
(Hi h d f) t l t bilit(High degree of) tool portability
Portability of skills
And, yes, even original authors think it’s a chimera, 
© Frank Leymann
i.e. they don’t think it’s a beautiful language
11
BPMN 1.x Filled a Gap
BPEL focused on language syntax and operational 
semantics
It sacrificed time‐to‐market and ignored visual 
representations by willp y
When looking back this was a BIG mistake!
BPEL turned out to be too difficult for mere mortalsBPEL turned out to be too difficult for mere mortals
BPMN 1 x filled this gapBPMN 1.x filled this gap
Non‐IT experts could communicate about business 
processesprocesses
h l
© Frank Leymann
Big progress in BPM technology
12
Chasm between BPMN 1.x and BPEL
BPEL processes are modeled with their future 
execution in mindexecution in mind
This has not been often the case in the past for BPMN
But more and more, BPMN users wanted their 
processes to be executed
Wide‐spread support of BPEL in workflow engines 
required to map BPMN to BPEL
But BPMN and BPEL metamodels are quite different
Thus mapping is not straightforward even complexThus, mapping is not straightforward, even complex
Many of you understand this much better than me
bl b d h h ?
© Frank Leymann
Problem: How to bridge this chasm?
13
Options to Bridge the Chasm
Standardize a visual representation for BPEL
Can be done as proven by the many graphical tools forCan be done as proven by the many graphical tools for 
BPEL
But BPMN contains a bunch of constructs withoutBut BPMN contains a bunch of constructs without 
BPEL counterparts
Corresponding BPEL extensions are requiredCorresponding BPEL extensions are required
Very time consuming!  (see BPEL4People…)
B t k f l ti !But users ask for a solution asap!
This was not practical
Thus, way chosen is to make (significant subset of) 
© Frank Leymann
, y ( g )
BPMN 2.0 more compatible with BPEL
14
Most Important Additions of BPMN 2.0
Clear operational semantics
“Nati e” XML e change format“Native” XML exchange format
Event & exception handlers (BPEL compliant)
Choreography extensions 
… and:
Pattern‐based mapping to BPEL (subset of BPMN)
Effectively, BPMN 2.0 contains a significant subset 
h l d fisomorphic to BPEL l Visual rendering of BPEL!
© Frank Leymann 15
The Price to Pay
Complexity of BPMN 2.0 has increased significantly!
Well ensuring executability is always complex!Well, ensuring executability is always complex!
Only a subset of BPMN 2.0 can be mapped to BPEL in 
a straightforward manner
It is still possible to model BPMN processes that have no 
canonical representation in BPEL
Some features have no runtime environment yet
E.g. corresponding domain is out of scope of BPEL at all
See BPEL4Chor
New middleware has to be developed
© Frank Leymann 16
Resulting Viewpoints
Viewpoint 1:
S bset of BPMN 2 0 is isomorphic to BPELSubset of BPMN 2.0 is isomorphic to BPEL
This subset is canonically transformed to BPEL and can be 
executed in BPEL enginesexecuted in BPEL engines
Corresponding subset is a visual layer on top of BPEL
Viewpoint 2:
BPMN 2.0 is a process language with well‐defined 
operational semantics imprinted by BPELoperational semantics imprinted by BPEL
It is possible to build a BPMN 2.0 engine without the 
need to use a separate BPEL engine
© Frank Leymann
need to use a separate BPEL engine
17
Confusion Ahead!
Now we have two process languages in the market 
supported by major vendorssupported by major vendors
Very likely, we will be faced with the situation that we 
will have both, BPEL engines and BPMN engines in 
the market
But what does it mean: BPEL engine? BPMN engine?
© Frank Leymann 18
Major Building Blocks of a Process Engine
Buildtime
Modeling Tool
Process Model
Process Engine
(aka Workflow System)
Workflow/
Process
Database
Runtime
(aka Workflow System)
Runtime
© Frank Leymann 19
Applications,…
More Detailed View: The WfMC Reference Model
Business Modeling
Workflow Enactment 
Engine
Administration,
Auditing
Workflow Enactment 
Engine
Worklist ApplicationWorklist
Processor
Application
Invocation
Worklist
Application
Application
© Frank Leymann
Modern Terminology
Process Modeling
Tool
Process
(Execution) Engine
Management
Tool
Process
Engine 2
Workitem ApplicationWorkitem
Management
Application
Invocation
Worklist
Application
Application
Function
© Frank Leymann
Yet More Modern Terminology
Process Modeling
Tool
Orchestrator
Management
Tool
Orchestrator 2
WorkitemWorkitem
Management
Service Bus
Worklist
Application
Service
© Frank Leymann
Architectural Impact of BPEL4People
WFMS WFMS Task Manager
Navigator
(aka Task Processor 
aka 
Human TaskNavigator
Workitem
Human Task 
Infrastructure)
Navigator
Manager
Past Architecture Today’s Architecture
© Frank Leymann
Architectural Impact: Refined View
WFMS
WFMS
Navigator Navigator WI
Status
Workitem
Protocol
Handler
WI
Status
Manager
Task
Protocol
Handler
Task‐Type‐Specific 
Interface
Workitem Interface Task
Manager
Task
Status
Worklist
Client
k
Task Client Interface
© Frank Leymann 24
Task List
Client
Resulting SOA‐View on WfMC Reference Model
Process Modeling Tool
Conceptual Process Model
Logical Process Model
Process
Engine
Management
Tool
Process
Engine 2
T k M S i BTask Manager Service Bus
© Frank Leymann 25
Task List
Client
Service
Process Modeling is Layered
Process Management Stack Data Management Stack
Conceptual Process Model
(BPMN)
Conceptual Data Model
(Entity-Relationship)(BPMN)
Transform
(Entity-Relationship)
Transform
Logical Process Model
(BPEL)
Logical Data Model
(SQL)( )
Deploy
( )
Deploy
Process Engine
(Internal Metamodel)
DBMS
(Relational Datamodel)
© Frank Leymann 26
DBMS: What Happened in the Past
Conceptual Data Model
(Object Model)
Data Model
Logical Data Model
(SQL)
Data Model
(Object Model)
DBMS
(Relational Datamodel)
OODBMS
(Object Model)
Data Model
(Object/XSD Model)(Relational Datamodel)
Conceptual Data Model
(Object/XSD Model)
E t d d RDBMS
Logical Data Model
p
(XSD)
Data Model
(XSD)
Extended RDBMS
(OR-XML Model)
Logical Data Model
(SQL)
XML-DBMS
(XSD)
© Frank Leymann 27
DBMS
(Relational Datamodel)
(XSD)
The Open Question
Will history repeat in the “Process World”?Will history repeat in the  Process World ?
© Frank Leymann 28
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 29
The Notion of a Native Metamodel
What does it mean to be a BPEL (BPMN) engine?
Well, the engine can import process models specified inWell, the engine can import process models specified in 
BPEL (BPMN), instantiate, run, monitor,… them according 
to the operational semantics of BPEL (BPMN)
Most often, during import BPEL (BPMN) artifacts will be 
mapped onto internal artifacts of the engine anyway: 
Most engines implement their own
i t t d lproprietary process metamodel
(Native Metamodel)
“Native” BPEL (BPMN) engines are rare exceptions!
N i BPEL (BPMN) BPEL (BPMN) i h i l d l
© Frank Leymann
Native BPEL (BPMN):= BPEL (BPMN) is the internal metamodel
30
What Determines an Engine’s Native Metamodel?
What the engineers building the engine chose to 
ensure efficiency, scalability etc of the execution!y, y
Modeling language(s) to be supported at time of 
creation of first release of process engine has deep p g p
impact on its native metamodel 
Good engine engineers take care about generality g g g y
and extensibility of native metamodel striving 
towards “straightforward” support of extensions of 
original modeling language(s) and even new modeling 
languages
Typically, users of a BPEL engine are not aware of the 
engine’s native metamodel and how different it is 
f BPEL t d l it lf
© Frank Leymann
from BPEL as a metamodel itself
31
The Role of Deployment
Logical Process Model
Conceptual Process Model
Model Import
Logical Process Model
WFMS
Deployment
Transformation
Engine
© Frank Leymann 32
Deployment
Deployment = Putting process model into production
I k it d f tiI.e. make it ready for execution
Deployment typically translates the corresponding 
model data into a different format e gmodel data into a different format, e.g.
Imported model might be specified in a metamodel 
d ff f h h d ldifferent from what the WFMS supports directly
See next!
…plus many other things happen
© Frank Leymann
p y g pp
33
Deployment: Key Aspect
<process …>
<sequence>
…
Navigator
…
<flow>
…
</flow>
Model Cache Instance Cache
Navigation
Logic
</sequence>
</process>
Logic …
Deployment
Activity Control Link Activity Instance
Process Engine Database
Activity Control Link Activity Instance
Name AID JCond … Source Target TCond …
PIID AIID AID …
…
© Frank Leymann 34
Native Support of a Metamodel
Each workflow engine supports a particular metamodel 
nativelyy
Typically, the native metamodel supported is…
reflected in the database of the WFMS
database schema holds instances of metamodel constructs 
immediately
database schema is tuned towards supporting the navigator optimallypp g g p y
reflected in the state model of the WFMS
all metamodel constructs have a set of states and transitions between 
themthem
state model is reflected in monitoring model and audit trail
directly implemented in the navigator of the WFMS
navigator understands each metamodel construct directly, their states 
and valid transitions, the relation between the states of different 
artifacts
© Frank Leymann
this implementation is “optimal”
35
Transformation During Deployment: Example
BPEL Tool
BPEL
Parse file and store artifacts
into internal optimized tables
FDL
Tables
into internal optimized tables
Workflow Enactment ServiceWorkflow Enactment Service
© Frank Leymann 36
Transformation During Deployment: Example
BPEL Tool
BPEL
Transform to proprietary
metamodel
WPDL
Workflow Enactment Service
Store artifacts into
internal tables
WPDL
Tables
Workflow Enactment Service
© Frank Leymann 37
Transformation During Deployment: Example
BPMN Tool
BPMN
BPEL BPEL
Workflow Enactment Service
Tables
Workflow Enactment Service
© Frank Leymann 38
Model Transformations
Process models specified in a different metamodel 
than the engine’s native metamodel may bethan the engine s native metamodel may be 
transformed into this native metamodel supported by 
a particular WFMSa particular WFMS
This way, a WFMS supporting WPDL natively may 
execute a BPEL model after corresponding 
f f ltransformation, for example
This way, a WFMS supporting BPEL natively may 
execute a BPMN model after corresponding 
© Frank Leymann
p g
transformation, for example
39
Problems of Model Transformations
Such transformations are not always faithful, i.e. the 
transformed model has different…
semantic behavior than the original model…semantic behavior than the original model
E.g. behavior of the source model has to be “emulated” in 
the target model – if possible at allg p
E.g. BPEL exception behavior is hard to emulate
…operational behavior than the original model
E.g. the navigator performs the transformed model less 
efficient than a WFMS supporting the metamodel of the 
original model
E g supporting FDL data flow in BPEL is cumbersome
© Frank Leymann
E.g. supporting FDL data flow in BPEL is cumbersome
40
Supporting One Metamodel Within Another
Metamodel M1
(BPEL)
Metamodel M1++
(BPEL++)
Metamodel M2
(BPEL)
Activities
(BPEL )
ActivitiesMetamodel M
(BPMN 2.0)
Activities
Link
Flow
ct t es
Link
Flow
Activities
Control Flow
Data Flow
Assign
Link
dataLink
Choreos
??? T l
BPEL‐D
Inefficient!
??? Topology
BPEL4ChorImpossible!
© Frank Leymann 41
Extending a Given Metamodel
Assume that models specified in a particular metamodel M2
must be “best” supported in an engine with a different 
M1metamodel M1
Th t t f M2 th t h d t l t i M1 tThen, constructs of M2 that are hard to emulate in M1 must 
be identified and corresponding constructs must be added to 
M1
This is why BPEL is designed to be extensible, to allow new 
constructs to be added to BPEL to facilitate optimal mappings 
from different metamodels
That way, an extended variant of a given engine (an M1++
engine) may actually support process models specified in
© Frank Leymann
engine) may actually support process models specified in 
different metamodels M2,M3,…, Mn
42
The Dawn of BPMN Engines
BPMN Tool
BPMN
2.0
BPMN BPMN
Workflow Enactment Service
2.0 Tables
Workflow Enactment Service
© Frank Leymann 43
…and it Might Get Surprising ☺
BPEL Tool
BPEL
BPMN BPMN
Workflow Enactment Service
2.0 Tables
Workflow Enactment Service
© Frank Leymann 44
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 45
Consequence For BPMN 2.0 Engines
Whether the BPMN 2.0 process model is executed by a 
BPEL engine or by a new BPMN 2.0 engine is not 
iimportant
Native metamodel of a BPMN 2.0 engine is likely to be 
different from BPMN 2 0 as a metamodel anyhowdifferent from BPMN 2.0 as a metamodel anyhow
Native metamodel of a BPMN 2.0 engine might be one 
that supports also BPEL today
I.e. vendor of a BPEL engine that provides import 
capabilities for BPMN 2.0 process models might become a 
vendor of a BPMN 2 0 engine based on the very samevendor of a BPMN 2.0 engine based on the very same 
execution engine
It is absolutely valid that an intermediate BPEL process 
model from the BPMN 2.0 process model is generated 
before the actual import takes place
users might not even be aware of this intermediate BPEL
© Frank Leymann
…users might not even be aware of this intermediate BPEL 
format at all
46
What is Important About a Process Engine?
Far more important than the native metamodel of a 
process engine (BPMN 2 0 engine too) is itsprocess engine (BPMN 2.0 engine too) is its 
robustness, efficiency, scalability etc
Providers of today’s BPEL engines typically have 
invested a lot into these non‐functional properties of 
their engines
Thus, such vendors have the capabilities to also offer , p
BPMN 2.0 engines with the very same non‐functional 
properties
© Frank Leymann
p p
47
A DBMS Analogy
In the past, new upcoming vendors implemented 
dedicated object oriented database systems OODBMSj y
The object model was natively supported by these DBMS
But robustness, efficiency, scale,… was often lacking 
by such systemsby such systems
Over time established RDBMS vendors supported keyOver time established RDBMS vendors supported key 
constructs of the object paradigm while providing 
robustness, efficiency, scale,…robustness, efficiency, scale,…
OODBMS are no mainstream anymore
© Frank Leymann
OODBMS are no mainstream anymore
48
Extrapolation to Process Engines
BPMN 2.0 is significantly more complex than BPEL
Thus vendors will subset BPMN 2 0 based on theirThus, vendors will subset BPMN 2.0 based on their 
customers’ requirements
It is likely that no vendor will support all of BPMN 2 0It is likely that no vendor will support all of BPMN 2.0 
in its product
BPEL engines will get extended over the time toBPEL engines will get extended over the time to 
support key features of BPMN 2.0 that are missing in 
BPEL and BPEL engines todayg y
BPEL engines will evolve to support relevant BPMN 
2.0 features (those missing in BPEL today)( g y)
So, relax and wait until today’s process engines 
support missing BPMN features (but robust, efficient, 
© Frank Leymann
pp g ( , ,
scaling,…)
49
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 50
How Execution is Enabled
Many execution relevant information is only available 
via the BPMN 2 0 schemavia the BPMN 2.0 schema
I.e. the graphical notation by far does not contain the 
information required to instantiate a BPMN process 
model and execute it
The corresponding syntactical details are given in the 
BPMN 2.0 spec as UML Class Diagrams and XML p g
Schema Definitions
But the given operational semantics
© Frank Leymann
But the given operational semantics
does not match this level of detail
51
Data Context In Processes
<import importType=”http://www.w3c.com/XMLschema”
location=”CustomerInfo.xsd”
namespace=”http://example.omg.org/Sample”/>
<itemDefinition id=”customerStructure” structureRef=”sd:customer”/>
<dataObject id=”CustomerDataObject”<dataObject id= CustomerDataObject
name=”Data Object to store Customer Information”
itemSubjectRef=”tns:customerStructure”/>
© Frank Leymann 52
Messages In Processes
<itemDefinition id=”tasksvcCustomerMessageStructure”g
structureRef=”tasksvc:MyFirstTaskServiceRequestMessage”/>
<message id=”myServiceCustomerMessage”
name=”MyFirstTaskService Request Message”name= MyFirstTaskService Request Message
structureRef=”tns:tasksvcCustomerMessageStructure”/>
© Frank Leymann 53
Mental Model About Task Input/Output
<serviceTask name=”Task” id=”Task” … >
…
<ioSpecification>
<dataInput id=”dataInput” …/>
<dataOutput id=”dataOutput” …/>
</ioSpecification>
<dataInputAssociation>
<sourceRef>InputDataObject</sourceRef><sourceRef>InputDataObject</sourceRef>
<targetRef>dataInput</targetRef>
</dataInputAssociation>
<dataOutputAssociation>
< R f>d t O t t</ R f><sourceRef>dataOutput</sourceRef>
<targetRef>OutputDataObject</targetRef>
</dataOutputAssociation>
…
© Frank Leymann 54
</serviceTask>
Transforming Data Objects to Task Data
<serviceTask name=”myTask” id=”myTask” … >
…
<ioSpecification>
<dataInput id=”myTaskInput” …/>
/<dataOutput id=”myTaskOutput” …/>
</ioSpecification>
<dataInputAssociation>
<assignment>
<from>bpmn:getDataObject(‘InputDataObject’)/XPath</from>
<assignment>*
| <t f >?
<assignment>*
| <t f >?<from>bpmn:getDataObject(‘InputDataObject’)/XPath</from>
<to>bpmn:getDataInput(‘myTaskInput’)/XPath</to>
</assignment>
<sourceRef>InputDataObject</sourceRef>
<targetRef>myTaskInput</targetRef>
</dataInputAssociation>
| <transform>? | <transform>?
</dataInputAssociation>
<dataOutputAssociation>
<transformation> expression </transformation>
<sourceRef>myTaskOutput</sourceRef>
<targetRef>OutputDataObject</targetRef><targetRef>OutputDataObject</targetRef>
</dataOutputAssociation>
…
</serviceTask>
© Frank Leymann 55
Summary: Important Relations For Input/Output
.wsdl/msg via  itemDefinition
<serviceTask name=”Task” id=”Task” … >
…
<ioSpecification>
<dataInput /><dataInput …/>
<dataOutput …/>
</ioSpecification>
<dataInputAssociation>
< R f> </ R f><sourceRef>…</sourceRef>
<targetRef>…</targetRef>
</dataInputAssociation>
…
</serviceTask>
© Frank Leymann 56
Defining Interfaces
<itemDefinition id=”ServiceInputMessageStructure”
structureRef=”fl:MyFirstTaskServiceRequestMessage”/>
Imported .wsdl (or .xsd or…)
3. structureRef fl:MyFirstTaskServiceRequestMessage />
<itemDefinition id=”tasksvcResultMessageStructure”
structureRef=”fl:MyFirstTaskServiceResponseMessage”/>
<message id=”myServiceInputMessage”
name=”My Request Message”
structureRef=”tns:ServiceInputMessageStructure”/>
<message id=”myServiceResultMessage”…/>
2.
<interface id=”MyInterface”
name=”MyInterface”>
<operation id=”MyOperation”
name=”MyOperation”>
<i M R f>t S i I tM </i M R f>
1.
<inMessageRef>tns:myServiceInputMessage</inMessageRef>
<outMessageRef>tns:myServiceResultMessage</outMessageRef>
</operation>
</interface>
© Frank Leymann 57
Service Tasks
<interface id=”MyInterface”
name=”MyInterface”>
<operation id=”MyOperation”
name=”MyOperation”>
<inMessageRef>tns:myServiceInputMessage</inMessageRef>
<outMessageRef>tns:myServiceResultMessage</outMessageRef>
</operation>
</interface></interface>
<serviceTask id=”MyTask” name=”MyTask”
operationRef=”tns:MyOperation”>
<ioSpecification>
<d t I t id “D t I t” “D t t b d t i ” /><dataInput id=”DataInput” name=”Data to be send to service” …/>
<dataOutput id=”DataOutput” name=”Result Output” …/>
…
</ioSpecification>
…
</serviceTask>
© Frank Leymann 58
Early Binding of Interfaces
<process id=”ServiceTaskProecss” name=”My Process”>
<partnerRole id=”SvcProviderRole” name=”Provider of Service Implementation”/>
…
<collaboration id=”MyCollab” name=”Service Task Collaboration”>
<participant id=”ServiceProvidingPartner”
name=”Partner Providing Service Implementations”>
<partnerRolRef>tns:SvcProviderRole</partnerRoleRef>
<interfaceRef>tns:BillingInterface</interfaceRef>
<interfaceRef>tns:PaymentInterface</interfaceRef>
<endPointRef>tns:MyBillingEndpoint</endPointRef>
<endPointRef>tns:MyPaymentEndpoint</endPointRef>
</participant>
<participant id=”OrderProcess”
name=”Process using services”>
<processRef>ServiceTaskProcess</processRef>
…
</participant>
…
</ ll b ti ></collaboration>
<endPoint id=”MyBillingEndpoint”> e.g. an EPR </endPoint>
<endPoint id=”MyPaymentEndpoint”> e.g. an EPR </endPoint>
Endpoint of the implementation of the interface
© Frank Leymann 59
Multi‐Instance Task
<dataObject isCollection=”true” … id=”DataObject1″/>
<dataObject isCollection=”true” … id=”DataObject2″/>
<serviceTask name=”MyMIService” id=”MyMIService”>
An “internal” multi‐instance
data object for each element
1. <serviceTask name MyMIService id MyMIService >
<multiInstanceLoopCharacteristics isSequential=”false”>
<loopDataInput>MultInstServiceInput</loopDataInput>
<loopDataOutput>MultInstServiceOutput</loopDataOutput>
<inputDataItem id=”singleMemberInput”/>
data object for each element
of which a separate
instance of the task
is created
<inputDataItem id=”singleMemberInput”/>
<outputDataItem id=”singleMemberOutput”/>
</multiInstanceLoopCharacteristics>
<ioSpecification>
An element (i.e. “item”) of the
multi‐instance loop data input/output
2.
<dataInput id=”MultInstServiceInput” … isCollection=”true”/>
<dataOutput id=”MultInstServiceOuput” … isCollection=”true”/>
</ioSpecification>
<dataInputAssociation …>p
<sourceRef>DataObject1</sourceRef>
<targetRef>MultInstServiceInput</targetRef>
</dataInputAssociation>
<dataOutputAssociation …>
<sourceRef>MultInstServiceOutput</sourceRef><sourceRef>MultInstServiceOutput</sourceRef>
<targetRef>DataObject2</targetRef>
</dataOutputAssociation>
</serviceTask>
© Frank Leymann 60
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 61
Basic Principle
A (recursive) function [..] is specified that allows to 
map a subset of BPMN to BPELmap a subset of BPMN to BPEL 
This map is specified by describing:
1 [ ] f ll l BPMN k h b d1. [t] for all elementary BPMN tasks t that can be mapped to 
BPEL
2 [e] for all elementary BPMN events e that can be mapped2. [e] for all elementary BPMN events e that can be mapped 
to BPEL
3 [s] for all BPMN structures s that have a straight‐forward3. [s] for all BPMN structures s that have a straight‐forward 
mapping to BPEL
This constructively defines the BPMN subset that canThis constructively defines the BPMN subset that can 
be mapped, as well as the associated map [..] itself
© Frank Leymann 62
Mapping a Process
© Frank Leymann 63
Mapping Data
© Frank Leymann 64
Mapping Messages and Interfaces
© Frank Leymann 65
Mapping Correlations
© Frank Leymann 66
Mapping Tasks
<invoke name=”[Task-name]”
partnerLink=”[Task-serviceRef]”
portType=”[Task-operation-interface]”
operation=”[Task-operation]”>
=
Send
Task
=
None
Task
<empty name=”[Task-name]”>
</empty>
operation [Task operation] >
</invoke>
© Frank Leymann 67
Mapping Task Input/Output
© Frank Leymann 68
Mapping Events (1/2)
© Frank Leymann 69
Mapping Events (2/2)
© Frank Leymann 70
Mapping Sequences
© Frank Leymann 71
Mapping If‐Then‐Else
© Frank Leymann 72
Mapping Pick
© Frank Leymann 73
Mapping Flow
© Frank Leymann 74
Mapping Parallel
© Frank Leymann 75
Mapping While‐Loops
© Frank Leymann 76
Mapping Until‐Loops
© Frank Leymann 77
Mapping Multi‐Instance‐Loop‐Characteristics
© Frank Leymann 78
Mapping Subprocesses
© Frank Leymann 79
Mapping Handlers
© Frank Leymann 80
Mapping Boundary Events (1/4)
© Frank Leymann 81
Mapping Boundary Events (2/4)
© Frank Leymann 82
Mapping Boundary Events (3/4)
© Frank Leymann 83
Mapping Boundary Events (1/4)
© Frank Leymann 84
AgendaAgenda
Historical RemarksHistorical Remarks
BPMN Origin and Positioning
The Notion of “Native Metamodel”
View on Process EnginesView on Process Engines
Selective BPMN 2.0 Features
Executable BPMN ‐ BPEL Mapping
SummarySummary
© Frank Leymann 85
Takeaways
BPMN 2.0…
…is a significant extension of BPMN 1.x
Ch hi C ll b i C iChoreographies, Collaborations, Conversations,
New event types and modes,
Event subprocesses,…
allows to specify executable process models…allows to specify executable process models
…contains a subset nearly isomorphic to BPEL
Especially, “BPMN 2.0 engines” can be build
Process engines implement their own internal metamodel
Native BPEL engines are very rare!
So will be native BPMN engines!So will be native BPMN engines!
Based on BPMN 2.0 complexity, 100% compliant 
implementations will take far more than a decade (if they will 
hit the market at all)hit the market at all)
Do you know a 100% BPEL compliant engine?
In practice, most process engines will support both, BPEL and 
( b f)
© Frank Leymann
(a subset of) BPMN 2.0
86
End of DocumentEnd of Document
© Frank Leymann 87