Deploying a Petascale-Capable Visualization and Analysis Tool

April 15, 2010April 15, 2010

Purpose of the next three talks

• Detail the VACET activities to deliver a petascale-capable tool to the Office of Science community (and others)– Ensuring S/W is capable of processing tomorrow’s

data– Ensuring that S/W scales (Joule)– Software engineering & deployment– Providing infrastructure to support the community– Outreach to the community

• Detail the VACET activities to deliver a petascale-capable tool to the Office of Science community (and others)– Ensuring S/W is capable of processing tomorrow’s

data– Ensuring that S/W scales (Joule)– Software engineering & deployment– Providing infrastructure to support the community– Outreach to the community

VisIt: Delivering a petascale-capable visualization and analysis tool to the Office of Science, the DoE, and more

• Problem• Office of Science application scientists need tools for visualization

and analysis (exploration, confirmation, & communication)• Solution

• VACET has extended VisIt to deal with unique Office of Science problems, including data size, and deployed to the community

• This includes data with trillions of cells using 10K’s cores• Impact

• Many Office of Science simulation codes now use VisIt. 11 letters of support from SciDAC-funded groups for VACET review

• Large capability delivered in a cost effective manner

2000 2005 Fall 2006

Project started

2007

VACET enablesmulti-institution development

APDEC retires ChomboVis for VisIt. Repurposes $’s for math

2008

VACET enables VisIt to run on trillions of cells and 10K’s cores

Summer 2009

VisIt becomes first ever non-simulation Joule code.

Fall 2009 2010

GNEP/NEAMS choose VisIt due to VACET leadership

2008

Both NSF XD centers commit to supporting VisIt

SW repository has ~30 developers from >10

institutions

We studied isocontouring and volume rendering, looking at up to 4T cells.

Visualization of 2 trillioncells, visualized with VisIt

on JaguarPF using 32,000 cores.

Visualization of 1 trillioncells, visualized with VisIt

on Franklin using 16,000 cores.

We demonstrated that VisIt performs well on tens of thousands of cores with trillions of cells.

• Goal was to uncover bottlenecks on tomorrow’s data.

• Experiments varied over supercomputing environment, data generation patterns, and I/O pattern.

• Goal was to uncover bottlenecks on tomorrow’s data.

• Experiments varied over supercomputing environment, data generation patterns, and I/O pattern.

Outreach

• We have worked hard to deploy VisIt, through tutorials, user support, documentation, etc.

• Tutorials:

• We have worked hard to deploy VisIt, through tutorials, user support, documentation, etc.

• Tutorials:Event Location Date Attendance

SC09 Portland, OR November 2009 ~50

Vis09 Atlantic City, NJ October 2009 ~75

NUG 2009 Boulder, CO October 2009 ~30

ACTS Berkeley, CA August 2009 ~35

Princeton Princeton, NJ July 2009 ~30

SciDAC 2009 San Diego, CA June 2009 ~40

CScADS Snowbird, UT July 2008 ~30

SciDAC 2008 Seattle, WA July 2008 ~40

SciDAC 2007 Cambridge, MA June 2007 ~20