Linking Abstract Plans of Scientific Experiments to their - - PowerPoint PPT Presentation

Linking Abstract Plans of Scientific Experiments to their Corresponding Execution Traces

Milan Markovic, Daniel Garijo*, and Peter Edwards University of Aberdeen *University of Southern California

Scientific Workflows

• Computational steps and data

dependencies that are necessary to carry out a scientific experiment

• Beneficial for reproducing

previous experiments, improving standardization practices in a research lab and educating students on existing methods

Workflows Abstractions

Current Limitations

• Complex plans are usually simplified into less detailed (abstract)

workflows

• Currently, it is difficult to link between high level workflow

specifications and low level execution traces

P-Plan & Prov-O

Execution Trace Bundle Plan wasDerivedFrom correspondsToVariable correspondsToStep correspondsToVariable

Input Data Sort Sorted Data

Extended P-Plan (EP-Plan)

• https://w3id.org/ep-plan
• New concepts for linking different abstractions of plans and execution

traces

• Concepts for additional plan metadata such as constraints, objectives,

agents, references to policies, and rationales for individual plan elements

• Intended for cross-domain applications

Linking Plans & sub-plans

• Multisteps are

decomposed into sub-plans

• Multivariables from

more abstract plans are linked to variables in sub- plans

:SummarizedWf (ep-plan:Plan) :ExecutedWf (ep-plan:Plan)

ep-plan:isDecomposedAsPlan ep-plan:isSubPlanOfPlan

:AggregateStep (ep-plan:Step) :File1Var (ep-plan:Variable) :AggregatedDataVar (ep-plan:Variable) ep-plan:hasOutputVariable :File2Var (ep-plan:Variable) :SortStep (ep-plan:Step) :OutputVar (ep-plan:Variable) :ErrorLogVar (ep-plan:Variable) ep-plan:hasInputVariabe ep-plan:hasOutputVariable ep-plan:hasInputVariable ep-plan: hasPart

:ExecuteWorkflowStep (ep-plan:MultiStep) :InputFilesVar (ep-plan:MultiVariable) ep-plan:hasInputVariabe :OutputFilesVar (ep-plan:MultiVariable) ep-plan:hasOutputVariabe

ep-plan:hasPart

ep-plan: isElementOfPlan

Linking execution traces

• MultiActivities

correpsond to MultiSteps

• EntityCollecions

correspond to MultiVariables

• hadMember relationships

link Entities and EntityCollections between different abstractions of execution traces

:SummarizedExecutionTrace (ep-plan:ExecutionTraceBundle) :ExecutionTrace (ep-plan:ExecutionTraceBundle) :SummarizedWf (ep-plan:Plan) :ExecutedWf (ep-plan:Plan)

ep-plan: isSubPlanOfPlan prov:wasDerivedFrom prov:wasDerivedFrom

:WorkflowEcexution (ep-plan:MultiActivity) :InputFiles (ep-plan:EntityCollection) prov:used :OutputFiles (ep-plan:EntityCollection) prov:wasGeneratedBy :Aggregate (ep-plan:Activity) :File1 (ep-plan:Entity) :AggregatedData (ep-plan:Entity) prov:wasGeneratedBy :File2 (ep-plan:Entity) :Sort (ep-plan:Activity) :Output (ep-plan:Entity) :ErrorLog (ep-plan:Entity) prov:used prov:wasGeneratedBy prov:used

prov: hadMember

prov:hadMember

ep-plan: isElementOfTrace

Future Work

• Implementation of EP-Plan vocabulary within the WINGS provenance

capture mechanisms

• Explore how other concepts supported by EP-Plan (e.g. references to

workflow constraints) can be captured within the WINGS platform

Contact & Acknowledgments

• milan.markovic@abdn.ac.uk
• dgarijo@isi.edu