155 South 1452 East Room 380
Salt Lake City, Utah 84112
1-801-585-1233
Integration of Reverse Monte-Carlo Ray Tracing
within Uintah
Todd Harman
Department of Mechanical Engineering
University of Utah
Isaac Hunsaker
Department of Chemical Engineering
University of Utah
This research was sponsored by the National Nuclear Security
Administration under the Accelerating Development of Retrofitable
CO2 Capture Technologies through Predictivity program through DOE
Cooperative Agreement DE-NA0000740
Deliverables
• Year 1: Proof-of-concept design for solving the RTE using
RMCRT within ARCHES/UINTAH.
• Year 2: Demonstration of a fully-coupled problem using
RMCRT within ARCHES/UINTAH.
Scalability demonstration.
• Year 3: Full burner scale V&V/UQ demonstration using
RMCRT.
Background
• Reverse Monte-Carlo Ray Tracing is an all-to-all method.
• All geometry information and radiative properties for the entire
computational domain must reside in each processor’s memory*.
Very restrictive
For example:
3 double precision arrays on a 500^3 domain = 3 Gbytes
* Xiaojing Sun. Reverse Monte Carlo Ray-Tracing For Radiative Heat Transfer in Combustion Systems. PhD
Dissertation, University of Utah, 2009.
Uintah
•
Use data from coarse regions/levels in the RMCRT calculations.
(3 double precision arrays on a 250^3 domain = 375 Mbytes)
Utilize the multi-level infrastructure in Uintah to project data from
fine to coarse levels.
•
What is a Level?
•
Collection of patches all with the same cell spacing.
•
Extents are determined by user or developer
•
Levels are overlapping.
•
Communication between levels with refining and coarsening
operators.
UCF: Inter-level Communication
Coarsen operator:
Mass/volume weighted or summation of fine data
UCF: Inter-level Communication
Refine Operator:
Approach
CFD: Always computed on the fine level.
RMCRT:
2 Level:
RMCRT on a coarse level, CFD on a fine level.
“Data Onion”: RMCRT & CFD on fine level, data pulled from
other coarse levels.
2 Level Approach
2 Levels
2 Level Approach
Advantages:
• Simple
•
Little to no heavy lifting, AMRICE, AMRMPM, AMRMPMICE
already using multi-level code.
Previous work suggests that this may be sufficient for pool
fires*.
•
Disadvantages:
• Accuracy (maybe).
Gautham Krishnamoorthy. Predicting Radiative Heat Transfer in Parallel Computations of Combustion. PhD Dissertation,
University of Utah, 2005.
Data Onion Approach
N-Levels
Data Onion Approach
1D
Coax UCF into mimicking a composite mesh that moves.
User inputs:
• Number of levels
• Refinement ratio between levels
• Step size
• Size of fine level patch
Data Onion Approach
Advantages:
•
Improved accuracy over 2 levels
Disadvantages:
•
Increased complexity
•
Coarsening data multiple times
•
More expensive.
Implementation Plan
Initial development in a light weight test-bed component.
Advantages
• Rapid testing of 2-level and data onion scheme.
• No disruptions to ARCHES users.
• Scalability studies focus on RMCRT tasks.
• Encapsulate verification code, no pollution in ARCHES
• Forces RMCRT tasks to be portable.
• Increased developer productivity from faster
recompile times.
Status: Completed
• Implemented a lightweight multi-level test-bed component.
• Implemented pseudo CFD and ray tracing tasks.
(RMCRT not quite ready)
• Implemented the 2 level scheme with pseudo CFD & ray tracing tasks.
• Implemented the data onion scheme, with pseudo CFD and ray tracing
tasks.
Status: Completed
•
Implementation of RMCRT tasks within the test-bed.
(single level and single patch)
•
80% Complete: Implementation of coarsening & refining code
•
75% Complete: Improve portability of RMCRT tasks.
•
50-70% Complete: Integration of RMCRT tasks within ARCHES
(single level and single patch)
Status: Pending
• RMCRT tasks with multiple patches.
• Verification of RMCRT tasks on a single patch.
• Scalability studies of RMCRT tasks on a single level.
• Implement 2-L RMCRT in test-bed.
Accuracy versus refinement ratio studies
Scalability Studies
• Implement 2-L RMCRT in ARCHES.
• Implement RMCRT with a data onion in the test-bed.
Accuracy versus number of levels, refinement ratio, size of fine patch.
Scalability Studies
Summary
On track for meeting year 1 deliverable:
“Proof-of-concept design for solving the radiative transport
equation using RMCRT within the ARCHES/UINTAH
framework.”