Project Topics 2012w

In groups of 1 or 2  students will complete a major course project on an approved topic. Groups may either develop a topic of their own, perhaps connected to their thesis topic and have it approved by the instructor or may selected from the list of suggested topics below.

• Pick a topic that interests you:
  • The best topics involve some algorithms desesign, an implementation, and a performance evaluation.
  • Pick a subject area that you know something about. Something to do with your thesis or interests in computing works well.
  • Do the one page write-up and discuss it with me.
  • Have a scan of the recent conferences and journals for something that interests you.
• Here are some topics that I am interested in and could give you some support with:
  • Cloud Performance: One of the challenges with moving parallel applications to the cloud is the wide range of storage options and their differing performance characteristics. Do a servey of cloud storage technologies and then an extensive performance benchmarking study. I can provide access to the Amazon EC2 Cloud. This project would require reasonable unix skills and an organized approach to benchmarking but not a lot of programming.
  • High Performance Storm Surge Modeling: I have a sequential Storm surge program written in Fortran and Matlab and would like to have it re-engineered to run on a cluster. This project would require you to either know or learn some Fortran.
  • Sparse Matrix Computation on GPUs: Many scientific codes can be reduced to BLAS (Basic Linear Algerbra Subprograms) routines. I am interested in Level 1 and 2 BLAS functionality running on GPUs. The idea behind this project would be to compare CPU and GPU codes for these operations written in C and Cuda. In particular, I am interested in accelerating R based codes that use these operations. This project would require some skill at programming close to the hardware level. You would need to learn Cuda along the way.
  • Read the book "Building Parallel Programs: SMPs, Clusters and Java" and pick a project based on using Java to do Cluster based parallel programming.
  • Monte Carlo Simulations on GPUs: Monte Carlo Style models seem on the surface well suited to GPU architectures. Design, implement, and evaluate a Monte Carlo Simulation framework for GPUs.