CS 258 Miniproject Part II

Todd Hodes <hodes@cs>

Second miniproject proposal:

In our first write-up, we introduced our initial Barnes-Hut design. There, we describe the framework and mention many of the interesting issues. For the remainder of the semester, we would like to investigate and implement the following:

Running on multiple architectures:

We will get our implementation running on some parallel machine other than the CM5. Preferably this would be the Cray T3D because that machine offers a different set of performance choices than the CM5 -- especially with regard to software caching. However if we are not able to get access to this machine, we will get the implementation working on some other parallel machine such as the HP cluster with Medusa cards, the Intel Paragon, or the NOW cluster over atm.

Software Caching:

Firstly, we would like to explore the relationship between software caching and execution speed. We will add software caching of node information to our implementation. We will then measure the speedup from the caching for both the tree-build and the tree-walk phases. We will also measure the additional memory required by the caching implementation.

Additional Parallel Tree-Build Algorithm:

Secondly, we would like to implement a second parallel tree-build algorithm. This algorithm could then be used to make fair comparisons to our existing parallel tree-build, and would additionally act as a testbed for the above caching strategies.

-----

maintained by <hodes@cs.berkeley.edu>