Next: Adaptive Poisson Solver.
Up: No Title
Previous: Monte Carlo Simulation.
These problems are offered by Professor Kathy Yelick, 2-8900,
yelick@cs.berkeley.edu.
- You can try to improve the ``pruning'' performance of
your TSP implementation by using another distributed data structure from
Multipol, like a hash table or trie, to keep track of paths visited
so far, to better prune paths which contain those paths.
- You can do a logic level device simulator (logic plus gate delays)
using the event graph in Multipol.
- Multipol contains an octtree data structure, which is used in the
Fast Multipole Method (FMM) or Barnes-Hut solution of the
N-body problem. An alternative octtree, which caches remote references,
is used by John Salmon for his N-body problem. Create a similar Multipol
oct-tree, and compare the performance on the rudimentary N-body solver
of Arvind Krishnamurthy.
- Implement a sophisticated N-body solver using an octtree.
Use it for Sharks and Fish 1+2 or 6, or ``bouncing balls with gravity.''
- Implement an existing PIC (``particle in cell'') code for
semiconductor device simulations in Split-C and Multipol.
- Add a sparse matrix class to Multipol, including the basic operations
(BLAS and level-0 preconditioning) mentioned during the talk by Roldan Pozo.
- Add an unstructured graph data structure to Multipol, including
operators to partition it, and iterators.
Use to to parallelize a finite difference application on an
irregular grid (EM3D). Also see if graph partitioning can be used to
reorganize a locally stored part of a graph to better exploit local
memory hierarchy.