toward interactive visualization in a distributed workflow steven g. parker oscar barney ayla khan...

29
Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize

Upload: lindsey-french

Post on 18-Jan-2016

218 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Toward interactive visualization in a distributed workflow

Toward interactive visualization in a distributed workflow

Steven G. Parker

Oscar Barney

Ayla Khan

Thiago Ize

Steven G. Parker

Oscar Barney

Ayla Khan

Thiago Ize

Page 2: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Component-Based Architectures

Experience with numerous component-based architectures

• CCA (Parallel, Method Invocation, multi-language)

• SCIRun (Shared memory, Dataflow, C++)• Uintah (Parallel, Method Invocation, C++)• Kepler (Single process + web services,

Generalized dataflow, Java +)• SCIRun2 (Distributed/Parallel, Multi-model,

mutli-language)

Page 3: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

DOE Common Component Architecture Project

A CA for large-scale Scientific Computation• Component Characteristics

May be SPMD or multi-threaded parallel objects• Heterogeneity

Parallel platforms to desktops and any language• Local and Remote

Parallel communication for remote parallel interfaces and 0-copy in-process connection

• Dynamic Composition and Integration Hot-swapable components, shared instances

• www.cca-forum.org

Open forum involving DOE labs, Universities, others

Page 4: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Uintah• CCA-ish component

architecture (C++ only)• Plus components for

multiphysics structured AMR simulations

• Scales to 2000+ processors

SimulationController

SimulationController

ProblemSpecification

ProblemSpecification

XMLSimulation

(One of Arches, ICE, MPM,

MPMICE, MPMArches

, …)

Simulation(One of Arches, ICE,

MPM, MPMICE,

MPMArches, …)

SchedulerScheduler

Tasks

DataArchiver

DataArchiver Tasks

Callbacks

CallbacksMPI

Assignments

LoadBalancer

LoadBalancer

Configuration

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.QuickTime™ and a

YUV420 codec decompressorare needed to see this picture.

Page 5: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SCIRun

Page 6: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SCIRun PowerApps: BioImage

Page 7: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

2/20/2004 Building a KEPLER Extension Using Ptolemy II

The KEPLER Systemfor Scientific Workflows …

A framework for design, execution and deployment of scientific workflows

Caters specifically to the domain scientist Builds on Ptolemy II Application pull from various projectshttp://kepler.ecoinformatics.org

Slide thanks to: Ilkay Altintas and Efrat JeagerSDSC UCSD

Page 8: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Kepler Workflow

Page 9: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Component Architecture Design Choices

• Degree of isolation: processes, threads, single address space?

• Mechanism for communication: dataflow, process networks, method invocation

• Synchronization• Programming languages: expressiveness tradeoffs• Data types explicitly supported• Performance requirements• Extra tools required?• Explicit support for parallelism?

Multiple designs for component architectures+ Tailored to application needs- Islands of functionality

Page 10: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SCIRun2• SCIRun2 provides a

component model for component models (metacomponents)

• Plug-ins provide support for:• CCA• SCIRun• Vtk• Others

• Components use “native” communication mechanisms to connect to similar components

• Bridges connect models

SCIRun2

Page 11: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Meta-components example

Common Framework

Driver Function

Integrator

Function

CORBACCA

Integrator

Driver Bridge

Page 12: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Application

Page 13: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SDM Requirements

Distributed Workflow• Repetitive• Shared resources• Automatically driven• Coarse-grained (seconds to minute per operation)

Interactive Visualization• Exploratory• Dedicated resources• User-driven• Fine-grained (milliseconds to seconds per operation)

Page 14: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Goal

What the user wants• To get work done• Make hard things easy

How to do this1. Combine tools with disparate strengths2. Make them work efficiently3. Focus on interfaces4. Enable consistent user interfaces

Page 15: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Utah's Contibution To the SPA Group

• SCIRun can now be controlled from SPA/Kepler workflows Server interface JNI interface

• “Smart” Re-run capability• Provenance framework

Page 16: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Kepler Workflow

Page 17: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Workflow Requirements and “Wants” We Address

• Seamless access to resources and services• “Smart” re-runs• Data provenance• Reliability and fault-tolerance• Detached execution

From: B. Ludäscher, et al. Scientific Workflow Management and the Kepler System. Concurrency and Computation: Practice & Experience, Special Issue on Scientific Workflows, to appear, 2005.

Page 18: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SCIRun With SPA/Kepler

• Kepler actor sends requests to a SCIRun server

• Useful for processing batch jobs or iterating through the parameter space of a SCIRun module (actor)

• Requires existing SCIRun network, which the workflow actor will tell SCIRun to load

• JNI interface to SCIRun

Page 19: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

SCIRun Server

• Simple TCP/IP server that can be started remotely by Kepler

• Accepts requests from client actor in the workflow and then sends back location of results when it has finished

• Allows for the possibility of remote or/and detached execution of SCIRun

Page 20: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

SCIRun and Kepler Dataflow Integration

Automate SCIRun network execution with a Kepler actor driving execution through a JNI interface or a remote connection to a SCIRun server

Incorporate SCIRun computation and visualization with the SPA workflow engine

Page 21: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

JNI interface with workflow

Page 22: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

What is provenance data?

• In general: steps taken to get a result• Information about computational

experiments or runs of scientific workflows that is needed to reproduce results

• We want to log metadata, steps applied to data, tools used to create data products

• Useful when you want to share/publish results

Page 23: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

The Standalone Provenance Frameworkhttp://kepler-project.org/Wiki.jsp?page=KeplerProvenanceFramework

Page 24: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

“Smart” re-runs

• Instead of running a workflow from scratch we only re-run parts of the workflow that have not been done before

• Example: we change a parameter downstream and dont want to re-run the actors that lead up to the one with the parameter change

• Especially useful in visualization pipelines and long running workflows

Page 25: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Utah and “Smart” Re-runs

• Uses VisTrails’ cache manager algorithm*• Idea is to re-run as little of the network as

possible by combining intermediate results from different workflow runs

• Recreates input to actors that need to be re-fired

* L. Bavoil, et al. VisTrails: Enabling Interactive Multiple-View Visualizations. IEEE Visualization, 2005.

Page 26: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Page 27: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

What is needed for “Smart” Re-runs

• We need to keep track of what we have done before

• Specifically we need to know what actors have been given what inputs with what outputs

• Stored provenance data can give us the information we need

Page 28: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Other uses for provenance data

• Recreate results• Recover from a system failure• Checkpoint a workflow• Create semantic links

Page 29: Toward interactive visualization in a distributed workflow Steven G. Parker Oscar Barney Ayla Khan Thiago Ize Steven G. Parker Oscar Barney Ayla Khan Thiago

Scientific Computing and Imaging Institute, University of UtahScientific Computing and Imaging Institute, University of Utah

Future work

• Continue work on “Smart” Re-runs system• Help workflow users integrate SCIRun with

their workflows• Get provenance framework checked into

Ptolemy CVS• Work on other provenance issues • Help SCIRun users take advantage of

workflow technology• Develop CCA to Kepler bridging mechanisms