Background

Some time ago, we got together with Terran Lane and Joe Kniss, faculty at UNM, to discuss potential SSDS collaborations we might explore.

Discussion centered on visualization for scientific applications

• projection, permutation, slicing, filters (for example, volume-rendered images)
  • ad-hoc strides on data, blocked for stride access but different viz needed
• pattern recognition
• tends to be bandwidth intensive, I/O bound
• compression vs. slicing
• ray-tracing, animation and rendering

What about randomized access pattern requests? Can we use a metabot to handle these in real-time or near?

Statistical summary in metadata is common

Neural imaging data (collaborations with MIND Institute)

• FMRI
  • scalar measurement over time/spzce
  • variable size data sets (~10GB)
• raw DTRI
  • tensor measurement
  • each file 2GB
• applications are multi-stage pipelines of transforms
• NIH might be interested in funding something if we can generate ideas