nadia ranaldo - eugenio zimeo
DESCRIPTION
Nadia Ranaldo - Eugenio Zimeo. Grids@Work 2008 ProActive and GCM User Group Orchestrating Services based on Active Objects and Grid Components. Department of Engineering University of Sannio – Benevento – Italy. Outline. Research Context Composition-based approaches for Grid applications - PowerPoint PPT PresentationTRANSCRIPT
Nadia Ranaldo - Eugenio Zimeo
Department of Engineering University of Sannio – Benevento – Italy
Grids@Work 2008 ProActive and GCM User Group
Orchestrating Services based on Active Objects and Grid Components
N. Ranaldo & E. Zimeo
Outline
• Research Context Composition-based approaches for Grid applications Service orchestration and choreography
• The SAWE Workflow Enactment System• Orchestration of ProActive/GCM components
Distributed data flow Dynamic binding
• Future directions
N. Ranaldo & E. Zimeo
Grid Applications: Composition-based approaches
•Complex scientific and business applications as composition of reusable, independent and cooperating software units in large-scale distributed systems
Heterogeneity, dynamicity, scalability, security, etc.•Composition in space
Structural relations and interactions among units Code re-use Tightly-coupled systems (closed world, well-defined
knowledge) Favoured by component-based architectures
•Composition in time Units ordered with respect to temporal dependences Efficient scheduling and resource usage Loosely-coupled systems (open world – incremental
knowledge, late binding) Favoured by service-oriented architectures
•Exploit the advantages of both the approaches for Grid applications
N. Ranaldo & E. Zimeo
Composition in time: Orchestration of Services
• Analysis
• Hypothesis
• Related work
• Propose experiments
• Define steps
• Prototype computing systems
• Perform experiments
• Data collection
• Visualization
• Validation
• Adjust experiment
• Refine hypothesis
• Presentation
• Dissemination
Define problems Experiments Data analysis Discovery
Graphical Workflow Editor
Workflow Engine (WE)
Experiment processes
Grid middleware functionalities
• scheduling
• data movement
• monitoring
N. Ranaldo & E. Zimeo
Workflow engines for e-science
Taverna:-Web services based language: Scufl;
-FreeFluo: engine
-Graphical viz of workflow
Kepler:-Actor,director
-MoML
-Execution models
-Ptolemy II
-Web Services
Triana:-Components
-Task graph
-Data/control flow
DAGMan:-Computing tasks
-DAG
Pegasus:-Based on DAGMan
-VDL
-DAG
and many others
N. Ranaldo & E. Zimeo
Towards Service Choreography: Centralized Orchestration Approach
Centralized control and data flow• Completely independent services• High network overhead
•A workflow is managed by a central workflow engine • Late binding• Efficient scheduling and QoS criteria fulfillment performed interacting
with resource management services (matchmaker, broker, etc.) and parallel execution frameworks (skeletons, parallel libraries, etc.)
W orkflowEngine
A1 A2
Client
activateA1
activateA2
An. . .
activateA n
data1 data2
datao ut
datan
data in
data in data1 datan -1
A1 A4A3 ... AnA2
W orkflowEngine
A1 A2
Client
activate A1
activate A2
An. . .
activate An
data1 data2
datao ut
datan -1
datan
data in
data in
Centralized control flow – distributed data flow• Dynamic dependencies among services
N. Ranaldo & E. Zimeo
Towards Service Choreography: Distributed Orchestration Approach
A1 A2Clientactivate A1 activate A2
An. . .
activate An
data1data2
datao u t
datandata in
• P2P network of services for discovery, composition and execution• Each activity described from the individual perspective of its
participating services• Better support to dynamic workflows
Re q u e stHan d le r
Exe cu to r
W E
G oaldescription
En viron men t description(seman tic an n otation s)
Co mp o se r
directlysolve
solve by com posinginner and external services
execu tepartia l w orkflow
Partia l Domainkn ow ledge
resu lt
N. Ranaldo & E. Zimeo
Semantic and Autonomic WE (SAWE)
• Compliant to WfMC specification• XPDL, BPEL• Configurator
Defines process description
• Engine Functional management
of the process
• Manager Monitors engine, running
activites, environment Decides actions to react
to events, environmental changes, etc.
N. Ranaldo & E. Zimeo
Workflows of ProActive/GCM Components
• A task is performed by a ProActive/GCM component (typically a composite component), which exports a well defined functionality
ProActive M iddle w are
Resources
Co
ntr
ol
Bin
din
g
SA
WE
Inte
rac
tio
n
Activ ity Control
. . .
Resource Inte rface
retrievesconcrete resource
executesactivity
schedules activity
Inv oker
X PDL Work flowDescription
Resource M anager
LocalR M
W SR M
H umanPerformer
R I
R M IR I
Jav aR eflection
R I
W ebServ ices
R I
ProActiv e/GC MR I
P rocess Control
Deploym entDescriptor/
ADL Definition
• Grid Component Model (GCM)• Based on Fractal• Target Grid context
• Parallel computation, deployment, dynamicity, autonomous behaviuor
• Lookup of already running components
• Deployment at run-time
N. Ranaldo & E. Zimeo
Future
Future
Engine
B
AA(run)A
Future
B(run)
Value
Value
B(block)B
Early-Start Pattern•Task anticipation exploiting asynchronous invocations and futures
Default future update strategy (data flow follows invocation flow)•Distributed data flow through futures
The lazy message-based update strategy
• No interactions among tasks and the engine for data updating
N. Ranaldo & E. Zimeo
Workflows of ProActive Components: Dynamic Binding
• Dynamic binding (abstract modelling) of ProActive tasks adopting the ProActive Scheduler
Co
ntr
ol
Bin
din
g
SA
WE
Inte
rac
tio
n
Activ ity Control
. . .
Resource Inte rface
ProActiveSche dule r
submits activity
retrievesconcrete resource
executeactivit ies
schedules activity
Inv oker
X P DL Work flowDescription
Resource M anager
LocalR M
W SR M
ProActiv eR M
H umanPerformer
R I
R M IR I
Jav aR eflect ion
R I
W ebServ ices
R I
ProActiv eR I
P rocess Control
retrieves result
Re s ource s
N. Ranaldo & E. Zimeo
Future Directions
• Distributed data flow based on the lazy message-based future update strategy
• Dynamic binding of ProActive/GCM components QoS description through semantic annotations of components
for dynamic binding based on user-defined QoS criteria
• Monitoring of ProActive/GCM components for autonomic behaviour of workflows
Thanks for your attention!
For further contact: Nadia Ranaldo [email protected] Eugenio Zimeo [email protected]