Table of Contents
Parallel SN: Lessons Learned
We are currently working toward the CY03 ASCI milestone
The road is long and tedious
We are in a transition phase
UCBL has been very useful, but has several fatal flaws
Early experience with true corner methods is encouraging
TSA is not robust for real applications
We are exploring a DSA option
Block-Jacobi works well for some applications, but not all
Speedup is linear!
We are exploring a parallel sweep option
Two level parallel strategy has worked well, but ...
It’s clear that MPI and OMP belong at the same level
We are in a transition phase
|