ParaView 3 Telecon 08-11-2009
| Item | People | Description |
|---|---|---|
| 1 | Moreland | Project Randy |
| 2 | Moreland | SC Tutorial |
| 3 | Moreland | Warning when ProcessRequests modifies itself |
| 4 | Geveci | Proposed changes to VTK pipeline |
| 5 | Moreland | Book omissions and errata |
| 6 | Utkarsh | Transition to git |
| 7 | Moreland | Stronger Dashboard
|
Project Randy
Previously:
Alan ran on Red Sky: 4K cores, 1.1 trillion cells. Alan applied Utkarsh's patches to completely remove the progress update, and that noticeable improved the performance. We should have an option to turn off progress reports when the number of processes exceeds some number, say 100.
Alan is still having problems scaling up past 5.5K cores or so, but this might be caused by the Red Sky system itself. Nathan reports similar problems using solvers without ParaView at all.
Today:
Ken got timings from Pat for running on Intrepid up to 32K cores. Pat has mentioned that might be some issues with the timings, possibly the job size being too big for the problem size.
In two weeks, Pat is going to Argonne and working with Vankate, who has been doing research on moving data quickly from computation platforms to visualization clusters. Going to integrate that into the coprocessing library.
We talked a bit about data staging. Everyone's doing it. Ken and Scott Klasky have had talked in the past about combining the paraview server with ADIOS data staging. Ken has an action item to send an email to Scott to perhaps organize a meeting at SC between Scott, Ken, and Berk.
News
Utkarsh and Berk are going to TACC next month. Work with Greg (something). Convinced them to use ParaView instead of VTK for the server side implementation.
Supercomputing Tutorial
- Ken Moreland, Andy Bauer, Nathan Fabian, and Pat Marion
- CoProcessing tutorial
Warning when ProcessRequest modifies itself
Previously:
A while ago we talked about adding a check to the executive that issues a warning if it detects that its MTime changes during a call to ProcessRequest. Whatever happened to that? I just spent 2 days tracking down a parallel pipeline update problem that would have been quickly alerted had this been in place (and was easily fixed once I put my own check in place).
Berk suggested first adding the check only to ParaView. We will see how that works, and then possibly move to VTK. Kitware will take care of that.
Today:
On Berk's todo list. Berk is working on proposals for the ARB. One is to change the modularization of VTK. The other is to yank out VTK 4.0 compatibility, which will allow us to clean up vtkExecutive change.
Book omissions and errata
Previously:
Lisa is going to create a bug-tracker like functionality for the Kitware books.
Transition to git
Branchy workflow. Tie in to stronger dashboard.
Install hooks in CMake? Marcus will be responsible for it.
Updates to the VTK submodule.
