Data Analysis Services Group - August 2012

News and Accomplishments

Accepted for publication: Clyne, J. Mininni, P., and Norton, A. Physically Based Feature Tracking for CFD Data, IEEE Transactions on Visualization and Computer Graphics, August, 2012

VAPOR Project

Project information is available at: http://www.vapor.ucar.edu

XD Vis Award:

• Fixed an issue for the animation of the probe texture, now the preview window and the VAPOR viewer are in sync when displaying IBFV textures.
• Began to transition the depth peeling prototype developed for VAPOR 2.0 to the current development release. Work also began on extending the Iso shader so that it can fully support depth peeling.

KISTI Award:

• John and Alan are working on a prototype visualization capability for ROMS data, a regional ocean model that was identified in the KISTI contract deliverables.  We are identifying the operations that must be performed in order to convert a ROMS dataset into a VAPOR VDC. John developed class objects that provide a high-level API for reading a collection of netCDF files representing a time-vary data set made up of 1D, 2D, and 3D variables. The new interface hides the underlying storage strategy options (e.g. multiple time steps per file vs one time step per file) from the caller.

• Alan made changes to the mom2vdf and momvdfcreate programs so that MOM and POP data can now be visualized using a stretched grid, and the missing values in the data can now be omitted from the visualization, producing much improved volume rendering.

Development:

• Ashish Dhital, a Siparcs intern from the U. of Wyoming worked with Alan to implement a keyframing capability for animation control in VAPOR.  Ashish completed his work before his departure at the start of August.  Since then, Alan has made changes to improve the user interface and improved the quaternion interpolation to use spherical spline quaternion interpolation (instead of Hermite interpolation).  The keyframing feature is now ready to release with vapor 2.2.
• John fixed a couple of Adaptive Mesh Refinement bugs that were cropping up with exceptionally deep hierarchies (e.g. 20 to 25 levels).

Software Research Projects

Feature Tracking:

Accepted for publication (finally!): Clyne, J. Mininni, P., and Norton, A. Physically Based Feature Tracking for CFD Data, IEEE Transactions on Visualization and Computer Graphics, August, 2012

NREL LDRD

A two-year NREL LDRD proposal submitted by former VAPOR intern Kenny Gruchalla, entitled Analysis and Storage Strategies for Large-Scale Wind Turbine Array Simulations, has been selected for funding. The VAPOR team are non-funded participants in this proposal to extend VAPOR's capabilities to better support wind turbine array simulations.

ASD Support

• Alan is providing visualization support for Baylor Fox-Kemper and Peter Hamlington’s ASD project that examines the turbulence in ocean water induced by contact with the atmosphere.  Peter has created data with a 10-fold increase in temporal resolution, which improves the time animation, however it appears that even higher temporal resolution is needed.
• John produced imagery for the Shay and Ireland teams to review. In the process of working with the Ireland data he discovered that the HDF5 output files were being written with "chunking", using a chunk size parameter that was giving very poor readback performance (e.g. ~20MByte/sec). After rewriting the data with a larger chunk size, readback performance increased to ~300MByte/sec. John also wrote a translator for Shay's data that would output Silo files. The Silo files are needed for parallel IO using VisIt, which is being used to visualize Shay's particle data.
• John also worked with David Richter, whose code employs a stretched computational grid. This work provided an opportunity to test out the recently developed stretched grid volume rendering code.
• Finally, John also met with Orlando Ayala's ASD team and provided them with a demo of VAPOR.
• Yannick created a python conversion script that allows VAPOR to render large, time-variant particle datasets with user specified geometry subdivision.

Production Visualization Services & Consulting

• Janice Coen is working on a visualization of the High Park fire using VAPOR.  Alan has been helping her with various visualization issues, and in the process identified a couple of bugs that were fixed.

Publications, Papers & Presentations

Accepted for publication: Clyne, J. Mininni, P., and Norton, A. Physically Based Feature Tracking for CFD Data, IEEE Transactions on Visualization and Computer Graphics, August, 2012

Systems Projects

Security & Administration Projects

• Continued tracking HUGAP project progress.  It is still not yet usable for production authentication and authorization purposes on GLADE resources. Ingested data from HUGAP to provide feed back to WEG about problems with data quality.  Started writing and testing rules to automatically generate passwd, shadow and group files based on this data and system specific data sources for the various classes of DASG servers.

System Monitoring Project

• Configured Nagios on glademgt1 to send remote host and service checks to CASG's Nagios server
• Turned on Nagios and Ganglia on glademgt1 in preparation for ATP and added metrics for infiniband performance.

NWSC Planning & Installation

• John worked with SSG to try to enable the graphics cards and remote rendering capabilities of both Geyser and Caldera. Efforts to do so thus far have failed, and IBM is investigating.
• John continued efforts to ready software for ATP testing of the DAV clusters Geyser and Caldera. After much effort John managed to build ParaView on one of the test system nodes, and then on Caldera itself. The build process was complicated both by the instability of the NWSC systems, as well as the size and inherent complexity of ParaView itself. The software has not yet been tested as the graphics capabilities of the DAV clusters have not yet been enabled.
• Created a working configuration for Xorg on booboo1 to allow testing of interactive graphics application.
• Continued working on picnic reconfiguration with picnicmgt2 as the xCAT master.
• Reinstalled picnimgt1 in order to boot into UEFI mode.
• Re-provisioned tnsd2 many times to try and debug the installation process and postscripts.  Found an issue where Redhat's Kickstart would wipe out all partition info from the DCS3700.
• Added accounts for vendor personnel on GLADE servers as needed.
• Installed the SGI Modular InfiniteStorage Server unit (delusion/riddle, named after Yellowstone lakes) that will provide NFS file systems for admin home directories and storage space for SSG use.  Started testing, found a bug in the Baseboard Management Controller (BMC) firmware that causes a hard BMC lockup when trying to access the GRUB boot menu via the Serial Over LAN (SOL) connection.  This prevents boot time actions, such as selection of an alternate kernel or booting to single user mode.  Opened a case with SGI for resolution. Started configuration and familiarization with the MIS Server hardware.  Asked SGI several times for more detaildd product configuration documentation to determine how to utilize the hardware efficiently.
• Tracked vendor installation and ATP progress reports.
• Participated in a Mellanox training video conference.
• Customized the IOR script for erebus to run in LSF environment.
• IOR and mdtest on erebus, IOR performance was ~600MB write with 1.1GB read on average from 8 node tests. mdtest never comleted (due to wallclock timeout), GPFS on erebus cannot handle the rate of metadata operation. What typically takes less than 20 minutes (on lynx with DVS) did not finish in 8 hour walltime.
• Traced the error messages on ibdiagnet outputs from NWSC glade. Noticed the verbsPorts mismatches, and failed uMAD datagram traffics and duplicate node descriptions.
• To troubleshoot IB traffic, built the GUID to LID and hostname table. This lookup file will be useful in tracing the outputs from the UFM servers.
  e.g. "ibtracert" will need LID numbers to generate the routing information and where errors are (specific ports of a Leaf or Spine module, or IB cables)

System Support

Data Analysis & Visualization Clusters

• Recompiled the netcdf package for Python-2.7.1 on mirage to fix an issue with a missing library.

GLADE Storage Cluster

• Added two new VETS systems as GPFS clients to mount Data01 and Data02.