Data Analysis Services Group - October 2011

VAPOR Project

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

TG GIG PY6 Award:

John and Yannick met with John Dennis to discuss steps needed to distribute and support the VDC extensions made to the PIO library. A decision was made to bundle the VDC support routines into the PIO distribution, enabling their use via a compilation option flag. A number of minor tasks were identified to be completed before the extended PIO library could be made publicly available. Yannick performed another round of benchmarking to compare compressed vs non-compressed data writes. The compression cost is not insignificant, but not unreasonable either (approximately double the time of uncompressed writes).

XD Vis Award:

Yannick modified  VAPOR's ShaderMgr to provide support for GLSL compilation pre-processing (e.g. #ifdefs). Numerous changes were required.

KISTI Award:

Minsu Joh visited  for two weeks to plan the KISTI work and also brought with her Sang-Myeong Oh, a grad student in atmospheric sciences with a lot of experience with the MOM4 ocean model.  Sang-Myeong with be with us through November and is working with the team to provide scientific input in phase 1 of the Kisti contract.  Alan worked with Sang-Myeong to convert his simulation output to vapor, using the scripts that Karamjeet Khalsa developed in August. Sang-Myeong ran a number of MOM4 simulations with various parameterizations to produce representative data sets. Alan worked with Sang-Myeong to identify his needs.  We found and
fixed a few bugs in the Python conversion scripts and in the VAPOR2.0.7 code.

Frank Bryan, a contributor on the KISTI award, was introduced to Minsu and Sang-Myeong, and brought up to speed on the project plans. Frank will continue to serve as a scientific adviser once Sang-Myeong returns to Korea.

Alan wrote a Python script to calculate the ocean depthat which a given temperature is reached.

Alan implemented a initial version of the momvdfcreate application that creates the VAPOR metadata associated with a MOM dataset.

Alan is in the process of creating the mom2vdf application, which will perform the conversion of MOM data to VAPOR.

John successfully prototyped a volume renderer and ray caster capable of handling both missing data, and terrain following grids. Both features are necessary for ocean data. He continued efforts to refactor VAPOR's internal data model to support abstract representations of regular, terrain following, and stretched grids. Once completed, the VAPOR code base will be greatly simplified, and will have greater flexibility for supporting less regular computational grids.

John explored options for supporting missing data values in VAPOR's wavelet transform library.

Outreach and Consulting:

Janice Coen provided Alan with new  fire simulation data
and is working with Alan to create illustrations of the data.

Alan worked with Mel Shapiro to create unsteady flow visualizations illustrating the particle trajectories associated with the ERICA storm of 1989, which was the most intense storm ever recorded south of 40 degrees latitude.  An animation we created of this is at http://vis.ucar.edu/~alan/shapiro/highETHUnsteady.mov

Michael Boettinger and Niklas Roeber of DKRZ visited for several
days, discussing some of the visualization efforts in their lab. Alan reviewed the VAPOR project with them, and they expressed a
particular interest in our ocean visualization effort.  DKRZ supports
an ocean model (MPI-OM) that may be possible to visualize with the same techniques we are developing for MOM and POP.

John worked with U.C. Berkeley's Tina Chow on visualizing idealized WRF simulations.

John continued efforts to organize a half-day visualization workshop for the AGU Fall meeting. All the speakers have been lined up, and most of the logistics for supporting the meeting are now in place.

Administrative:

John authored CISLAR sections on large data handling, and the VAPOR project, and authored the POPPR section on the Twister cluster.

Misc:

Alan and John attended visweek 2011 in Providence, RI, where some new and
interesting visualization techniques were presented.

Data Analysis & Visualization Lab Projects

File System Space Management Project

• Gave up on EMC being able to determine why Networker is backing up too much stuff.  EMC suggested reordering the timing of the incremental backups to run before the full backups to try to avoid the problem. Decided that an enhancement to the Filesystem Management Utility would be a suitable replacement for Networker-based backups that would be easier to deal with and present a superior end-user interface for restoring files.
• Continued work on FMU design and documentation.

Security & Administration Projects

• Started documentation on the steps required to configure and support a MyProxy certificate authority server.  Read up on X.509 certificates and the related RFCs.

CISL Projects

GLADE Project

• Created scripts to provide GLADE accounting records in a specified format from the output of Robinhood filesystem scans.
• Wrote a Perl script to easily start multiple threads of "dd" I/O tests across multiple nodes.
• Began work on a script to send the "mmfsadm vfsstats show" output to ganglia in order to graph GPFS system calls (read,write,open,close,getattr,etc.).
• Evaluated "mmfsadm" and "mmtrace" tools for better GPFS maintenance and tuning. Among many options in mmfsadm, identified "mmfsadm vfsstats show" to be useful for tracking performance.
• Evaluated presentations of EOFS(European Open File System) workshop held in September in Paris. Some of the troubleshooting examples will be useful in the future.

Lustre Project

• Installed the Virtual Box to test the Lustre2.1 on RHEL6 (current test setup is on RHEL5). Two virtual nodes were built to work as MDS+client and OSS server. Also compiled the packages from the source code and tested the custom made RPMs. Only unexpected obstacle was the need to downgrade the "automake" package to deal with the older Lustre configure script.
• Exchanged messages with Jazcek at CU to assess the Lustre performance and tuning options and scheduled a visit.

Data Transfer Services Project

• Continued testing MyProxy UCAS authenticated certificates with Globus Online and GridFTP to learn how this really works.
• Checked the globus-url-copy option for separate authentication for source and destination, which will be needed for Janus system at CU for possible batch mode transfers.  There is a separate "authorization assertion file" option usable for newer globus5 which is the current version on our gridftp servers.
• Responded to Erich's question on COS override for htar commands. Identified the override option and modified the wrapper script to conform to expeced COS.

System Support

Data Analysis & Visualization Clusters

• Recompiled NCO to include Netcdf4/HDF5 support
• Recompiled R to include tcltk support and install the extRemes package.
• Worked with NETS to resolve network connectivity issues with Mirage5 after the power down. Diagnosed mirage5 failure to reboot after powerdown. It does come back despite the virtual terminal hang. The hang was due to network connection being blocked, which was resolved after contacting NETS. The boot option was also changed to provide consistent virtual terminal display.
• Installed the soloii software on all systems.
• Responded to GPFS hang blocking a directory /glade/proj3/DART/raeder/T85x80. The root cause turned out to be a deadlock state job that was terminated without cleanup. After reboots on bluefire nodes that contain the locked processes, the file system came back to normal operation.  It took longer to identify due to scheduler not showing his processes. Next time similar hang happens, we will be able to check processes on bluefire nodes first and handle it faster.
• Responded to EVticket (68552) on Kevin Raeder's issue with slow "ls -l" causing his job failures.  Identified the location of the offending lines in the scripts and suggested a work-around for faster response.

GLADE Storage Cluster

• Replaced a failed fan module and 3 failed drives on the DCS9900.
• Provided DSS with a custom ganglia graph to show estimates of monthly data02 usage.
• Checked the /glade/home migration plan for bluefire and identified 670 users still with /blhome/ as default login.
• On Oct 11, /gpfs/blhome reconfiguration was done in coordination with SSG group.