Internal:ReportCloud

CBC3D

Procedure for Aneurysm meshes

1. High memory node on Turing acquired.
2. Activated the topology checks for the image and the mesh.
3. Generation of meshes for:
   • Stent
   • Aneurysm + Stent
   • Aneurysm
4. Aneurysm mesh:
   • Preprocessing for the initial image of the aneurysm to be acquired:
     • Combination of the slices to transform from 2D to 3D image.
     • Relabeled the areas of the image where the label was set to stent, now to the label of the aneurysm.
   • Run the image as input to CBC3D on Turing:
     • Parameters:

NumIterations	NonConnectivity	Fidelity	LatticeSpacing
7	3	0.95	0.15

Final: Extracted surface of the meshes and transformed to STL.

NRR

Preliminary Results

The MPI Driver is currently running on crtc-gpu.cs.odu.edu. It has run 3702 iterations of Media Case 17.

Current best results (better than HD=4.25 of Media paper):

HD	BlockSizeX	BlockSizeY	BlockSizeZ	WindowSizeX	WindowSizeY	WindowSizeZ	nonConnectivity	selectionFraction	numApproxSteps	numInterSteps	numAdaptiveIterations	zeroBlocksFraction	rejectionFraction
2.18	1	1	1	3	3	2	2	0.15	10	1	5	0.00001	0.2
2.44	1	1	1	3	3	2	2	0.10	9	0	5	0.00001	0.2
2.63	1	1	1	3	3	1	2	0.05	10	0	5	0.0001	0.35
3.19	1	1	1	3	3	1	2	0.05	9	1	5	0.001	0.2
3.30	1	1	1	3	3	1	2	0.10	9	0	5	0.00001	0.35
3.42	1	1	1	3	3	1	2	0.10	9	0	5	0.0001	0.25
3.57	1	1	1	3	3	1	2	0.10	9	0	5	0.0001	0.20
3.61	1	1	1	3	3	1	2	0.05	9	0	4	0.00001	0.20
3.83	1	1	1	3	3	1	2	0.05	9	0	3	0.0001	0.35
4.04	1	1	1	3	3	1	2	0.05	9	0	3	0.001	0.20

Current Work

The ongoing work and completed work includes:

1. Discussed PAPBNRR compilation with Angelos.
   • Angelos compiled everything successfully
2. Discussed the required script modifications for PAPBNRR with Angelos.
3. Locate process pinning within ITK (Fotis' version)
   • This had been temporarily resolved by reserving an entire node on Turing
   • Modifications to ITK to remove pinning are complete.
4. Focus is currently on reduction of intermediary temporary files
   1. Locate where files are first written and remove need for empty initial files
   2. Locate where files are read/used
   3. The goal is to keep structures in memory and
      1. pass the in-memory structures between PAPBNRR and Mesh generators (PODM, CBC3D).
      2. Implement converters to transform the data structures from PODM/CBC3D into the format used by PAPBNRR.
         • (Possible alternative) Based on Fotis' suggestion VTK structures might be replaced by ITK structures.
      3. keep data structures in memory between PAPBNRR iterations.
5. Completed:
   • Detailed installation instructions of NRR code (including ITK, VTK, CGAL, PODM, CBC3D)
   • MPI code to run multiple instances
   • Files-Memory:
   1. Files produced during execution of NRR:Found how to keep them only in memory, the files don’t need to be written - In progress
   2. Files passed between NRR (Image) and PODM (Mesh) - Conversion: In progress

Action Items

The current (i.e., immediate) action items include:

1. Review the scripts created by Angelos (Accessible Friday, 6 April, afternoon).
2. Explore feasibility of the gnu parallel utility within Angelos' SLURM scripts.
3. Modify MPI driver to archive output (mesh, deformation field, HD) for each job
4. Explore Python Multi-threaded driver

These items are short term expected to be complete end of day Monday, 9 April.

Cloud Compute Platforms

This section documents the current research into Cloud Compute services. Currently this includes only AWS.

AWS Reference Materials

• AWS Overview
• AWS Getting Started
• AWS Batch Jobs
• AWS Command Line